JP4255376B2 - Nonlinear distortion compensation for modem receivers. - Google Patents

Abstract

A method and modem for communicating serial input data over a transmission link are disclosed. Serial input data is partitioned into parallel data elements prior to rotation by an invertible linear mapping. Computationally efficient multi-rate wavelet filter banks are employed in a receiver for analyzing a received signal. Self-interference components are calculated in the receiver per sub-band and tap-weights for the filter banks are derived as a function over the range of self-interference values. A weighted sum is formed and subtracted from the delayed, received signal to eliminate the interference, and is removed from the recovered signal.

Description

[Cross-reference with related applications]
Not applicable

[Statement on Federal Aid or Development]
Not applicable

[Background of the invention]
As is known in the art, a modulator-demodulator (modem) is an electronic device that modulates a transmitted signal and demodulates a received signal. This modem generally provides an interface between the digital device and the analog communication system, thus allowing analog transmission of digital information between two terminals or stations. Such transmissions are typically made through transmission links such as band-limited telephone lines, cellular communication links, satellite links, and cable TV. That is, information is transmitted through the transmission link only in a predetermined range of frequencies having the maximum bit error rate.

  As is also known, a modem is used to provide wireless transmission between the transmitting station and the receiving station. Such wireless communication can be accomplished using VHF, IS-54 (mobile), IS-95 (mobile), SPADE (satellite), GSM (mobile) using one of the linear modulation techniques QAM, QPSK, Pi / 4DPSK, and GMSK. It can be used for various applications such as HDTV, SAT-TV. Depending on the wireless communication application, each bandwidth is limited within an acceptable bit error rate.

  Most modems can compensate for Gaussian noise, but cannot manage impulse noise well. Most modems also require higher power amplification means since amplitude distortion is unacceptable in all cases. In the broadcast environment, it is understood that FM transmission provides an excellent response to impulse noise. Also in the FM transmission scheme, the transmission signal is amplified with an efficiency of almost 100%. This is because the information carrying the portion of the signal is identified by the zero cross point of the signal. Thus, amplitude distortion is ignored.

  Despite the advantages of these devices, there are only a few independent examples of nonlinear frequency modulation used in modems. Frequency shift keying giving a data rate of 300 bps uses essentially two different frequencies to represent independent data bits. The next major development in modems is the use of phase modulation, starting with two-phase modulation and progressing to four-phase and eight-phase. A combination of amplitude and phase modulation was later developed and is also referred to as quadrature amplitude modulation or QAM.

  Subsequent development is Gaussian minimum shift keying or GMSK. At first glance, such encoding is similar to four-phase modulation, but to avoid amplitude modulation, a special low-pass filter called a Gaussian filter is applied to the data input to the phase modulator. Since Gaussian filters are considered to be similar to integrators, it is also argued that such modulation schemes cause frequency modulation. This is because GMSK can be demodulated using an FM discriminator. Still, GMSK applies a linear function of data to the in-phase and quadrature carriers to produce a linear modulation. Nevertheless, true FM is mathematically equivalent to applying the trigonometric function of the filtered signal to in-phase and quadrature carriers. FM is a non-linear modulation, which is considered insufficient for data transmission. This is because the transmission frequency spectrum is not a simple conversion of the baseband spectrum as in an AM modem. Another possible problem with FM for a modem is that only the true signal is accepted by the modem's voltage controlled oscillator. That is, the equivalent in-phase and quadrature carrier method for FM is redundant when compared to the double in-band and quadrature carrier double-sideband spectrum of a QAM modem, and is therefore considered inefficient. Both paths transmit the same data generated in the double sideband spectrum. Also, single sideband (SSB) transmission has been considered undesirable for modems. This is because there is no simple technique for efficiently modulating an AM-SSB signal without a carrier reference. FM as an analog technology that has nothing to do with data transfer has been forgotten, except for FSK.

In a typical modem, binary data is typically passed through a baseband high order cosine filter. This filter limits the bandwidth of the baseband signal. Therefore, when multiplied by the carrier to the baseband signal, the control of the passband signal bandwidth is made without intersymbol interference. A typical modem output includes a signal with discrete phases. For this reason, the data contained therein are identified by recognizing the phase of each bit. For example, whenever a signal has a + 90 ° phase shift, it is interpreted as zero. When the phase shift is -90 °, it represents 1. Thus, in a typical modem using a carrier, the phase and / or amplitude of the carrier signal is determined by the symbol currently being transmitted. This carrier assumes only selected values of phase and amplitude for most of the duration of each symbol . Also, all graphical plots of selected phase-amplitude pairs are called modem constellation points. A typical modem requires that there be a distinct point in the constellation for each possible value of the transmitted symbol . Furthermore, if these points are assigned erroneously, due to intersymbol interference, or due to the link-like noise, bit error occurs at the receiver.

  In many applications, the computational requirements of the modem introduce a delay that is detrimental to the operation of the system. For example, digital voice transmission and multi-access networks are sensitive to delays in modems. Furthermore, the rate per unit of bandwidth at which the modem transmits and receives data is referred to as modem bandwidth efficiency. In the discipline of digital information theory, this efficiency is known to be maximized when the transmitted signal has maximum entropy or randomness. Maximum entropy transmission is band-limited Gaussian noise, and among other properties, Gaussian noise does not exist in distinctly different phase-amplitude pairs as in constellations. Thus, it is desirable to provide a modem with a passband carrier that minimizes internal processing time and maximizes bandwidth efficiency without sacrificing bit error rate.

Method for communicating symbols sequential series through transmission link, the transmission comprising the steps of polyphase filters the symbol in a multi-rate, the method comprising modulating a carrier using the output of the filtered, modulated carrier and transmitting over the link, thresholding and reassembly against receiving a modulated signal, and applying the inverse function of the polyphase filter of the transmitter to the received signal, the output of the inverse function filter and a step of recovering the transmitted symbol subjected to a treatment.

  Data is transmitted from the first modem to the second modem over the wireless transmission link. In this case, an input data frame is formed from the input data, the input frame is multiplied by the rotation matrix, the output of the rotation matrix is frequency-modulated and transmitted, the transmitted data is received, the frequency is demodulated, and the demodulated data Is multiplied by the second rotation matrix, and the reversely rotated data is reassembled to recover the original data.

A modem that transmits symbols through a transmission link includes a transmitter and a receiver. The transmission unit includes a division element for dividing the input into parallel data channels, a baseband transmission rotation unit for multi-phase converting the channel-divided data into parallel signal channels, and the parallel signal channels into sequential serial samples. A rearrangement unit for conversion; a carrier modulator for providing a modulated signal; and a transmitter for transmitting the modulated signal. The receiver unit receives a modulated signal transmitted thereto, a demodulator that demodulates the received signal into a parallel signal channel, and multi-phase converts the received and demodulated signal into a parallel data channel A receiving rotator and a built-in element for incorporating the parallel data channel into the serial data signal.

These methods of multi-phase filtering and this modem allow FM modulation of transmitted symbols . This is because only the true component of the original symbol is generated. FM provides enhanced immunity to non-Gaussian noise, regardless of whether it is achieved by frequency or phase modulation of the input signal, uses non-coherent IF, and thus does not require carrier recovery, It is cheaper than conventional modems by the absence of an A / D converter, provides low co-interference due to FM capture effects, and is compatible with analog signals by using interchangeable operators. Amplifiers that are power efficient but potentially non-linear, such as class B or class C, can be used. This is because the zero cross point can be used to determine data content and carrier recovery is not required. Modems that use modulation schemes such as QAM cannot use non-linear amplification as such. In addition, satellite modems using the methods and modems disclosed herein save TWT backoff power and are thus more energy efficient. This is because personal modulation achieves higher data rates without sacrificing bit error rate, while intermodulation is not a problem.

  In the first embodiment, the multiphase filter processing is performed by a pair of wavelet type filters (for example, orthogonal mirror filters). The dividing element divides serial data in a plurality of parallel data channels prior to linear phase FIR vector filtering. In this case, the filter coefficient is a square matrix. Thereby, the input data is converted into a parallel signal channel. This conversion is by convolutional rotation of the input data vector. Each coordinate of the output signal is limited to a frequency subband that slightly overlaps its neighbors. Prior to the rotation of the first embodiment, the transmitter pre-emphasis places most of the information in a low baseband frequency range. This is due to the noise probability density function of the FM discriminator proportional to the square of the frequency. Receiver de-emphasis results in an addition to the overall gain equation. This equation includes, in one embodiment, FM transmitter gain contribution, de-emphasis gain, and noise reduction gain. The pulse amplitude levels representing the divided data bits in each subband do not necessarily need to correspond to integer bits as long as all levels in all subbands correspond to integer values.

The receiver portion provides a reverse rotation filter for performing an inverse transform that recovers the original data. In one embodiment, the inverse transform is interchangeable with the modulation transform. In another embodiment, the inverse rotation filter coefficients of the receiver are adaptively selected for equalization to compensate for channel distortion. This is because the analyzer is a partial rate FIR filter. Thus, a near perfect reconstruction filter is used. The threshold operator uses the nearest integer coordinate value as the most probable symbol .

  In another embodiment, interchangeable rotation and counter-rotation filters are derived from a base matrix describing the geometric rotation of the vector. These functions are for converting an input data vector in the data coordinate system into a signal vector in the signal coordinate system. As a result, the sequentially serialized coordinates of the signal vector form a digital sample of the band limited analog signal. Yet another way to work with matrices is a mathematical transformation with a discrete wavelet transformation.

  In each of the transmitter and receiver portions, the ideal rotation operator is a computationally efficient multi-rate wavelet filter bank. Logarithmic amplification of the baseband signal prior to the introduction of the FM transmitter into the modulator improves the modulation gain of the receiver. Furthermore, noise introduced into the transmission channel is attenuated as a by-product of logarithmic amplification preceding the transmitter and inverse amplification following the receiver.

Dispersive impairments within the link create a relative phase shift between subbands. In this case, the “orthogonality” of the wavelet is lost, and cross terms appear as self-interference in the recovered symbols .

[Brief Summary of Invention]
In the invention disclosed herein, the cross term in the approximation to order the differencing phase shift of a specific, fixed tap weight filter is used at the receiver. The output of this filter is weighted by a known function F of the measured differencing phase shift (p), further filter output weighted is subtracted from the recovered signal to remove the self-interference. In a preferred design, fixed coefficients of the filter is calculated from the dot product of one value of differencing phase shift. If the phase cannot be measured, an adaptive combiner is used.

  The foregoing features of the invention, as well as the invention itself, will be more fully understood from the following detailed description of the drawings.

Detailed Description of the Invention
Referring to FIG. 1 , the signal decomposition-resynthesis system 10 includes an analyzer 12 having one input port 14a and a pair of output ports 16a, 16b. Each of the output ports 16a, 16b is coupled to a corresponding pair of input ports 18a, 18b of the synthesizer 20.

Analog input signal X supplied to the input port 14a of the analyzer 12 is decomposed into a pair of signals W _'1 and V' _1, respectively provided to the output port 16a, 16b corresponding port. Similarly, the input port 18a, a pair of input signals W _'1, V' ₁ given to 18b synthesizer 20 is reconstructed in the output signal Y at the output port 20a of the synthesizer 20. The decomposition and reconstruction process performed by analyzer 12 and synthesizer 20 is further described below, as are signals V ′ ₁ and W ′ ₁ . Here, it is noted that the input signal X is decomposed into signals V ′ ₁ and W ′ ₁ , but the signals V ′ ₁ and W ′ ₁ will later be reconstructed to ensure that the input signal X is reconstructed. Combined.

It should be noted that the building block analyzer 12 and synthesizer 20 operate digitally, but for the sake of clarity the analog and digital signals are required to be converted to each other in the system. The signal conditioning circuit is omitted. Such a signal conditioning circuit is described below in connection with FIG. However, in short, and as described below in connection with FIG. 4, when an input signal to a system building block such as analyzer 12 or synthesizer 20 corresponds to an analog signal, the signal is It should first be fed through a filter having a filter characteristic selected to provide Nyquist filtering. The appropriately filtered signal is then sampled by an analog to digital converter (ADC). Similarly, if the signal from the system building block is an analog signal, the signal is fed to a digital / analog converter (DAC) and has a filter characteristic that is also selected to provide Nyquist filtering. 2 filter.

The synthesizer 20 and the analyzer 12 perform inverse function operations of each other. That is, when the signals W ′ ₁ and V ′ ₁ from the analyzer 12 are applied as input signals to the input ports 18 a and 18 b of the synthesizer 20 , the output signal Y is equal to the sequence of the input samples X except for a predetermined delay time. Become. In the preferred embodiment, this delay time corresponds to one sample time.

Similarly, when the signals V ′ ₁ , W ′ ₁ are applied to the respective input ports 18 a, 18 b of the synthesizer 20 and the resulting output signal Y is applied to the input port 14 a of the analyzer 12 , the respective outputs of the analyzer 12. The original signals W ′ ₁ and V ′ ₁ are obtained at the ports 16 a and 16 b.

As described above, analyzer 12 and synthesizer 20 provide signal decomposition and reconstruction functions. As described below, the analyzer 12 and synthesizer 20 can be used as building blocks and can be combined to provide more complex circuits. These circuits are combined to provide various signal transmission / reception systems. Furthermore, the analyzer 12 and synthesizer 20 can be implemented efficiently by hardware, software or a combination of hardware and software, as well as other system building blocks described below.

The analyzer 12 and synthesizer 20 described herein operate according to an ordered sequence. This sequence is a sample from, but not limited to, an analog to digital converter (ADC). For example, the sample is expressed as X (0), X (1), X (2), X (3). In this case, X (0) is the latest sample. Each of the binary numbers X (0) to X (3) has a special value within a predetermined value range. For example, an 8-bit ADC gives a range of decimal values from -128 to +127.

  The sequence of samples X (0) to X (3) can be considered as the coordinates of the vector [X] in the four-dimensional space. It is possible to perform linear conversion from the X coordinate system to another coordinate system in the same four-dimensional space. Thus, vector X is converted to vector Y by “rotation” matrix C. In matrix notation, this is expressed as:

  Using the mutually orthogonal axes of the new coordinate system, the vector Y has components corresponding to the projection of X onto the new transformation axis. Such a projection is given by forming a dot product (inner product) of the vector.

  For example, in the case of four dimensions, a set of Walsh codes Code1 to Code4 is given as follows.

  The Walsh codes Code1 to Code4 represent four-dimensional orthogonal coordinate axes of the time order space. Codes Code 1-4 have a length corresponding to the square root of 4 (ie, the dot product of Code 1 and itself is equal to 4), and thus are not unit vectors.

  The rotation matrix C is expressed as follows.

That is, the rows C _{1 to} C ₄ of the rotation matrix C correspond to the components of the Walsh code vector. In this case, matrix multiplication of the rotation matrix C and the vector X is equivalent to a dot product of a special one of the vectors C _{1 to} C ₄ and the vector X. The components of the row vector X are expressed as follows:

  The new state vector is represented as follows:

Here, ^* is a symbol indicating dot product operation between [X] and Walsh code vectors [C ₁ ] to [C ₄ ]. The new state vector [Y] completely describes the original state defined by [X] and is calculated for each group of 4 samples X (0) -X (3).

  The linear rotation operation can be reliably reversed and has the following relationship.

  Here, [C ^] is an inverse function of [C]. The matrix C ^ is obtained from the following relational expression.

  Here, L corresponds to the dimension of the row or column vector of the matrix C ^.

In the two-dimensional case, the Walsh vector is represented as C ₁ = [+ 1, + 1] and C ₂ = [+ 1, −1]. Thus, the analyzer 12 is defined to have two outputs using the above two-dimensional vector.

The Walsh vector is generated by the two-dimensional vectors C ₁ and C ₂ in any dimension. That is, replacing a set of Walsh codes with a two-dimensional generator matrix gives a new Walsh code with twice the dimensions. Using this procedure, the N-dimensional transformation necessary to provide the analyzer 12 is obtained. Thus, because of the approach in which Walsh vectors are constructed from their two-dimensional Walsh generators, higher dimensional analyzers 12 and synthesizers 20 can be constructed from the two dimensional case.

  This matrix transformation method generates a matrix equation for providing a modem or signal encryption device. The basic operations in the calculation required are addition and subtraction of terms (terms).

It should be noted that in the case of the cascade analyzer shown in FIG. 1A or the cascade synthesizer shown in FIG. 1B given using a matrix method with a matrix having N dimensions, the lowest frequency channel is determined by the matrix vector C ₁ . Each subsequent channel i is computed by the corresponding matrix vector C _i and finally the highest frequency channel is computed by the matrix vector C _N. Thus, in the case of 4-dimensional matrix, the lowest frequency channel is calculated by the matrix vector C _1, and the highest frequency channels is calculated by a matrix vector C _4. The case of tree analyzers and tree synthesizers provided by this matrix method is described below with reference to FIGS. 1C and 1D, respectively . The analyzer 12 is used as a sub-analyzer in cascade analyzers and tree analyzers. The synthesizer 20 is also used as a sub-synthesizer in cascade synthesizers and tree synthesizers.

In another preferred method, equations describing the output signals V ′ and W ′ of the analyzer 12 are given as follows:

  Here, V 'corresponds to the scaling or filtering function of the sampled input signal X. W 'corresponds to the residual or derivative of the sampled input signal X. X (n) corresponds to the latest input sample. X (n-1) and X (n-2) correspond to two preceding input samples. SHIFT corresponds to a set of variables equal to a positive power of 2 (eg, 32, 64, etc.). BN also corresponds to a positive integer having a value between 0 and SHIFT, and is preferably a relatively small value.

  As described in more detail below, the scaling function V 'and residual W' defined by equations Equations 1 and 2 are evaluated on alternating input samples of the input signal X. In this case, the value of the residual W ′ is expressed as “center” sample X (n−1) and X (n−1) and X (n−2), which are closest to X (n−1). This corresponds to twice the difference between the midpoint of the line connecting the two represented samples. When equations 1 and 2 are evaluated with alternating input samples, only odd (or even) “center points” are selected and calculated. Thus, the output rate of the sub-analyzer corresponds to ½ of the input rate of the input signal X.

  The residual W 'can also be interpreted as a deceleration around the center point. Thus, in Equations 1 and 2 above, it is possible to define the residual W 'as acceleration and replace W' with -W '. This provides an alternative essentially equivalent representation for the residual W '.

  Other implementations that use closer neighbors and define the residual W 'as the higher or first derivative of the function evaluated for the center point can also be used.

  The interpretation of the residual W '(n) described above indicates that the residual W' (n) responds only to the slope change around the alternating samples. Thus, one characteristic of the residual W 'in the present invention is that the residual W' is given a zero value when the slope about the center point is constant.

  The normally defined residual W (n) has the following relationship with the effective residual W ′ (n) defined in the present invention.

  As a result of the definition of the residual W ′ (n) of the present invention, a signal based on linear or non-linear quantization of the residual W ′ (n) (rather than the residual W (n) defined in the normal solution) Compression results in improved performance over compression schemes. This is based on the usual definition of residual W (N) as shown below.

  The operation of the subsynthesizer is described by the following equations Equations 3 and 4.

  Where SHIFT corresponds to a power of 2 and

It is. Y ′ (−) corresponds to the pre-calculated and stored (ie, recursive) Y ′ (n) value.

  Practical implementation of these equations by defining the variable SHIFT as above and taking advantage of the fact that multiplication and division by powers of 2 are equivalent to right or left shift on binary computers Is done by relatively small and simple hardware and is therefore preferred.

  Thus, in the solution of the present invention, the residual W 'is defined to give a special feature and the sequence V' is defined to give the remainder of the sequence.

Referring to FIG. 1A , a so-called “cascade analyzer” 24 includes a plurality of, in this case, N sub-analyzers 24a to 24N . The cascade analyzer 24 applies the signal V ′ ₁ from the output port of the first sub-analyzer 24 a to the input port of the second sub-analyzer 24 b, and supplies the signal V ′ ₂ from the output port of the second sub-analyzer 3rd. This is given to the input port of a sub-analyzer (not shown) and so on. This process continues until the sub-analyzer 24N provides a signal V ′ _N having a predetermined sample rate.

For example, in the communication system, as described above, it is desirable to combine a plurality of, for example, N sub-analyzers. In this case, _N is selected so that the signal V ′ _N selectively has a sample rate that is less than or equal to twice the lower frequency cutoff of the communication link. Each sub-analyzer halves the sample rate at its input, so the various output signals W ′ ₁ , W ′ ₂ ... W ′ _N of the cascade analyzer are given at different rates.

Referring to FIG. 1B , a so-called “cascade synthesizer” 26 is provided by a plurality of sub-synthesizers 26a-26N of the type described in connection with FIG . This cascade synthesizer operates to provide the inverse function operation of the cascade analyzer.

Referring to FIG. 1C , a so-called “tree analyzer” 28 is provided by appropriately combining a plurality of sub-analyzers 29a-29g, each of the type described in connection with FIG . Thus, in addition to the cascade emanating from the scale function V ′ (N), it is possible to have the remaining cascade emanating from one or more remaining sequences W ′ (N). That is, the remaining sequences W ′ (N) can be considered as inputs for multi-resolution analysis. Therefore, in that case, it is possible to give a scaling function V ′ and a remaining sequence W ′ with equal sample rates. Here, the sample rate corresponds to 1/8 of the original sample rate.

  The tree analyzer 28 is provided by analyzing each of the remaining sequence outputs W 'until all such sequences are dropped to one rate. This rate is usually below the Nyquist rate of the lower cutoff frequency of the transmission link with which the analyzer cooperates. The three-level analyzer tree 28 is given here to have all of its outputs 29a-29h at 1/8 the sample rate r of the analog input signal X. Each output sample can have more bits per sample than the input samples, but the number represented by those output samples is usually small in size and is easily recreated by a quantizer (not shown). Quantized.

Referring to FIG. 1D , a so-called “tree synthesizer” 30 is provided by a plurality of sub-synthesizers 31a-31g, each of the type described in connection with FIG . The tree synthesizer 30 operates to provide the inverse function operation of the tree analyzer 28 ( FIG. 1C ).

It should be noted that in each of the applications described below in connection with FIGS. 2-8, the term analyzer used through 8 represents a sub-analyzer, cascade analyzer or tree analyzer element, etc. It is used for. The term synthesizer referenced in FIGS. 2-8 is used to indicate a sub-synthesizer, cascade synthesizer or tree synthesizer element. Each of these are those described in connection with FIG. 1 through 1d. Each of the analyzer and synthesizer elements is given by a matrix transformation technique. Instead, the analyzer and synthesizer elements are each given by a three-point equation that takes the form of Equations 1 and 2 above. Further, each of the analyzer and synthesizer elements is given by an equation of the form

If the analyzer and synthesizer are given as a tree-type analyzer and synthesizer, the matrix vector Ci operates on the bottom channel of the tree analyzer or synthesizer if i = 1, 3, 5,. On the other hand, if i = 2, 4, 6..., The matrix vector Ci operates on the upper half channel of the tree analyzer or synthesizer. Thus, if the analyzer 28 having a channel 29a~29h is given by eight-dimensional matrix having a vector _C 1 -C _8, channel 29h~29e is computed by a matrix vector _{C 1,} C 3, _{C 5} and _{C 7,} the channel 29d~29a is calculated by a matrix vector _{C 2, C 4, C 6} and _{C 8.}

Referring to FIG. 2 , the signal encryption device 32 includes an analyzer 33 having a signal encryption circuit 34 coupled thereto, for example, a random number generator. This analyzer 33 is given by combining a plurality of, here five, sub-analyzers as shown. Each of the sub-analyzers is of the same type as the sub-analyzer described above with respect to FIG. 1 and thus operates in a similar manner. Each of the plurality of analyzer output ports 33a-33f is coupled to a corresponding one of a plurality of similar input ports 36a-36f of the synthesizer 36. Similarly, a synthesizer 36 is provided by combining a plurality, here five, of subsynthesizers as shown. Each of the sub-analyzers is of the same type as the sub-analyzer described above with respect to FIG. 1 and thus operates in a similar manner.

It should be noted here that five sub-analyzers and sub-synthesizers are connected as shown, but any number of sub-analyzers and sub-synthesizers can be used, as will be appreciated by those skilled in the art. That's what it means. It should also be noted here that, as described above, cascaded sub-analyzers and sub-synthesizers are shown here, but tree-type sub-analyzers and sub-synthesizers can also be used. Therefore, one input signal X is decomposed into an arbitrary number N of signals. In general, and as described above with reference to FIG. 1 , the decomposition procedure performed by the analyzer, and therefore the number of combined sub-analyzers, is such that the Nyquist frequency of the _Nth signal V _N is a known lower side of the input signal X. It is preferable to end the process when the frequency becomes lower than the cutoff frequency.

In a general operation overview of the signal encryption circuit 32, an input signal X, for example an audio signal, is supplied to an analyzer input port 33a 'and is decomposed by the analyzer 33 in the manner described in connection with FIG. . Here, the input signal X is decomposed into signals V ₁ , V ₂ , V ₃ , V ₄ , V ₅ and W ₁ , W ₂ , W ₃ , W ₄ , W ₅ as shown. In order to encrypt, a random number generator 34, which is a digital random number generator, supplies an encrypted signal to each of the signal paths of the residuals W ′ ₁ to W ′ ₅ . The respective values of the residuals W ′ ₁ to W ′ ₅ are thus corrected to correspond to the signals E _{1 to} E ₅ .

As shown here, one method for encrypting a residual W _{'1 ~} W' _5, the logical variables having a value corresponding to the sign and a logic value of the residual W _{'1 ~} W' ₅ Gives an exclusive-or (XOR) operation (as shown). Instead, another encryption method is provided by adding a plurality of individual secret binary bit streams to one or all of the residuals W ′. In general, the residual signal W ′ is provided to have a relatively low power level. Thus, residual signals W ′ ₁ -W ′ ₅ appear as simple noise (ie, “buried in” noise) signals by either or both of the encryption techniques described above. Other encryption methods known to those skilled in the art can also be used. For example, channel permutation and symbol substitution. It should be noted here that each of the signals V ₁ ′ to V ₅ ′ may be encrypted instead, or a combination of the signals W _K ′ and V _K ′ may be encrypted instead. is there.

  It should be noted that the signal encryption device 32 can be modified to include digital cryptographic feedback that is common in the field of encryption. In order to provide digital cryptographic feedback in the present invention, each of the five XOR outputs from the analyzer is in the form of a data encryption standard (DES) technique for encryption, and five independent secret random number generators (not shown). ) Should be fed back to the input.

  It should also be noted here that means for providing digital random numbers well known to those skilled in the art, such as, but not limited to, DES can be used with or without cryptographic feedback. In addition, other methods of finding and using random numbers to modify the value of the residual W 'can also be used, such as, but not limited to, additional noise techniques that do not use an XOR function.

Signal _E 1 to E _5, which is encrypted, is then coupled to a corresponding one of the output ports 33a~33f the analyzer, then supplied to the synthesizer input port 36 a to 36 e, also unmodified signal _{V 5} is input Supplied to port 36f. Synthesizer 36 uses the equation Equation3 and 4, to reconstruct the signal _E 1 to E ₅ supplied thereto, give thereby reconstructing an output signal Y. This signal Y is transmitted through a communication channel (not shown), for example.

Receiver (not shown) receives the signal transmitted, further analyzes the signals to recover the signal V ₅ and E ₁ to E _5. The receiver then recovers the signal _W 1 to _W-5 decodes the signals _E 1 to E _5, obtain the original signal X and further re-synthesis.

The synthesizer 36 reconstructs the signals E ₁ -E ₅ in this case so that the reconstructed signal X ′ provided to the synthesizer output port does not occupy beyond the Nyquist bandwidth. It is. Thus, it is possible to make the sampling frequency of the highest frequency encryption stage correspond to twice the upper cutoff frequency of the transmission channel.

  The reconstructed encrypted signal X 'applied to the synthesizer output port will of course have a different (noisy) power spectrum and a different average total power due to the additional noise introduced. The input signal power is reduced to limit the total channel power. The means for limiting the total channel power includes means for transmitting the power factor to the receiver by using the amplitude of the pilot tone used to synchronize the decryptor and the ADC. These pilot tones are used, for example, to determine the ADC clock by a phase locked loop and are given as explicit narrowband tones. Alternatively, these pilot tones are added to the residual W 'and provided as a secret coding sequence to hide the true value of this residual. The receiver correlates with the appropriate residual to establish time synchronization. Certain coded sequences (eg, GOLD codes or JPL range codes) having a suitable correlation required to obtain synchronization by this method are known in the field of spread spectrum communication systems.

  Furthermore, the encryption device reliably recovers the signal from the receiver to approximately the same extent as the original input signal. However, there are noisy variations due to noise from the channel and inaccurate time synchronization of the ADC converter in the end of the transmitter and receiver.

The signal encryption apparatus also includes means (not shown) for line equalization. Such means are well known to those skilled in the modem design arts. Line loss changes with frequency shift and phase shift are compensated by adaptive filtering. Often, it has a precursor burst of known energy for setting the filter parameters and means for periodically modifying the coefficients of the equalizer based on a constant quality of the received signal.

  It should be noted that when the value of the input signal X is set to zero in the signal encryption device 32, the output signal X ′ of the signal encryption device is generated by random noise generated by the encryption random number generator. It corresponds to that. Thus, in this case, the signal encryption device 32 acts as a broadband random number generator. Thus, the signal encryption apparatus is used as a generator for providing a signal on a line used in a line equalization process.

  Furthermore, the analyzer is used to measure the channel response of each channel. Thus, when the synthesizer provides a test signal to each channel, the analyzer measures the channel response of each channel to determine the loss of each channel. In this way, the analyzer and synthesizer are used to provide a line equalization method.

As explained below in connection with FIG. 4 , this concept is further refined to use only a synthesizer at the transmitting end and only an analyzer at the receiving end.

  The delay in the tree (Delay) corresponds to the following equation.

  Here, L is the number of levels in the tree. Thus, the tree process of a five level tree introduces a relatively short delay of only 31 samples. Similarly, for cascaded systems, it is believed that the delay in each signal path is given by the above equation where L corresponds to the number of stages in the special path of the cascade. This short delay is a desirable feature because human hearing is sensitive to echoes from hybrid couplers found in most transmission media.

Referring to FIG. 3 , a transmission / reception system 40 for transmitting / receiving a high reliability signal includes a transmission unit 40a and a reception unit 40b. The transmission unit 40 a includes a signal adjustment circuit 41. The circuit preferably includes an input filter 42 having filter characteristics that are preferably selected to Nyquist filter the analog signal supplied thereto. This filter 42 couples the analog signal to an analog / digital converter (ADC) 43. The converter converts the received analog signal into a digital signal representing the analog signal.

  The analog signal supplied to the input port of the ADC 43 is supplied to the ADC 43 through an amplifier or other pre-processing circuit (not shown). Preferably, the signal preprocessing circuit, such as a low noise amplifier or a buffer amplifier, is characterized as a relatively wideband amplifier, and even having a relatively low level of phase dispersion over the band of the amplifier. That is, this amplifier gives a phase shift to the amplified output signal. This phase shift is substantially equal to the amplified output signal therefrom, at least over the bandwidth of the transmitted signal. Furthermore, it is desirable that the sampling rate of the ADC 43 is greater than twice the Nyquist sampling frequency (that is, greater than twice the frequency of the highest frequency component signal in the input spectrum).

  The ADC 43 converts the signal provided from the filter 42 according to a predetermined sampling rate, and provides a digital word stream. At the output of the ADC 43, such a stream of digital words is supplied to the signal encryption circuit 44.

The signal encryption circuit 44 combines the analyzer 46 appropriately decomposing the signal supplied thereto into a plurality of signals, the encryption circuit 48 encrypting the decomposed signal, and the signal supplied thereto. And a synthesizer 50 for making a reconstructed encrypted signal. The signal encryption circuit 44 is of the same type as the signal encryption circuit 32 described above in connection with FIG. 2 and operates in a similar manner.

  The signal encryption circuit 44 encrypts the signal by adding the secret number to the residual W ′. In this case, the secret number value is known only to the transmitting side and the receiving side. Such addition is typically made modulo-2. That is, by applying a bitwise exclusive OR (XOR) function to the residual signal. Such addition is also accomplished by simply adding modulo to the actual data size of the data.

  The encryption device 48 is provided as a secret number generator of a type that gives a different secret number based on the number of times it is required to do so, or whose output depends only on its input. The latter type is often self-synchronizing. The former type can lose synchronization between the encoder and decoder if the receiver loses bit or word synchronization with the transmitter for any reason.

The encryption device 48 provides permutation operations between multiple channels. Instead, the encryption device 48 provides a replacement operation. Alternatively, the encryption device 48 provides the logic or operation described in connection with FIG .

  The signal encryption circuit 44 supplies the reconstructed encrypted signal to the input port of the output signal adjustment circuit 51. This circuit has a digital / analog converter (DAC) 52. This converter is given as a logarithmic ADC and converts, for example, the encrypted bit stream supplied thereto into an analog signal representing this encrypted bit stream. This analog signal is then fed to an output filter 54 having filter characteristics appropriately selected as described above. This filter 54 couples the signal supplied thereto to a first end of a transmission channel 56 which is provided, for example, as a telephone line. The second end of the transmission channel 56 is coupled to the receiver 40b of the transceiver system 40.

  In general overview, the receiver 40b receives the reliable signal supplied thereto and decodes the signal to provide a clear text signal to the output port. The receiving unit 40b of the transmission / reception system 40 receives the encrypted signal through the input signal adjustment circuit 57 at the input port of the input filter 58 having an appropriately selected filter characteristic. Input filter 58 couples this signal to ADC 60. The ADC 60 converts the analog signal and provides a stream of digital words in the same manner as described above. The filter 58 is provided to have a low-pass filter frequency cutoff characteristic corresponding to 1/2 of the sampling frequency of the ADC 60. The ADC 60 supplies a stream of digital words to the decryption circuit 61. The decoding circuit 61 includes an analyzer 62 that decomposes a signal supplied thereto into a plurality of signals, and a decoding device circuit 64.

  When the encoding circuit 48 of the transmission unit 40a of the system encrypts the signal using the secret number, the decryption device circuit 64 subtracts the secret number to recover the original data.

The combination of the decryptor and synthesizer 64 , 66 performs the inverse function operation of the combination of the encryptor 48 and analyzer 46 at the transmitter, and the decrypted and appropriately recombined signal is converted to a digital / analog converter (DAC). ) 68. The DAC 68 provides an analog signal corresponding to the bitstream supplied thereto and supplies the analog signal to a receiver output filter having appropriately selected filter characteristics as described above.

  If the analyzer and synthesizer are given as a cascade or tree, as described above, the number of levels is typically determined by the lower limit of the input signal bandwidth. Thus, the actual requirement for most applications is to place a pre-filter (not shown) having high-pass or band-pass filter characteristics in front of the signal conditioning circuit 41.

  In any case, regardless of whether or not such a filter is provided, the transmitting and receiving systems will not recover signals with bandwidth that exceeds the bandwidth of the communication link 56. In applications where the communication link 56 is provided by electromagnetic or sonic energy, the bandwidth of the link typically has practical limits, such as a single television (TV) channel.

Referring to FIG. 4 , signal encryption system 72 receives an analog signal at an input port, filters the analog signal, and further through filter 74 and ADC 76 in the same manner as described above in connection with FIG. Convert to the first bitstream. This bit stream is supplied from the ADC 76 to the signal encryption device 78. The signal encryption device 78 receives these signals directly on the separated signal channel. A random number generator 80 is coupled to the signal encryption device 78 and provides a random bit stream to the signal encryption device 78. This random bit stream modifies the first bit stream to provide an encrypted signal. The signal encryption device 78 combines these signals and provides the reconstructed encrypted signal at its output port.

  The encrypted signal is then supplied from the signal encryption device 78 to the input port of the DAC 82. The DAC 82 converts the encrypted bit stream into an analog signal. This analog signal is coupled to the input port of the signal combination circuit 86 through a filter 84 that provides an appropriately filtered signal.

The timing circuit 87 has a timing signal generator 88 for providing a timing signal. The timing circuit 87 provides an analog or digital timing signal to the receiver 92. For this reason, an optional ADC 90 is here coupled between the timing signal generator 88 and the adder circuit 86. If the DAC 82 is included in the timing circuit 87 , the timing signal is provided to the digital input port of the DAC 82 through an optional signal path 91 '. If the ADC 90 is omitted, the timing signal is supplied to the adder circuit 86 through the signal path 91 as shown.

  It should be noted here that those skilled in the art will recognize that the inherent recovery of information at the receiving end of an analog communication link using an analyzer and synthesizer operating according to the above equation is: This means that it is necessary to synchronize the ADC of the transmission unit of the system and the DAC of the reception unit of the system so that the transmission unit and the reception unit of the system coincide with each other with a certain sampling time. A large inaccuracy in synchronization results in words that cannot be translated from the receiver. Small statistical jitter of timing synchronization has a similar effect as noise on the link.

Timing and synchronization schemes are known to those skilled in the art. For example, in one method, an oscillator locked to (ie derived from) the sampling rate of the transmitter is transmitted over the link. The received oscillator signal is used to obtain a receiver sampling clock signal (eg, by a phase locked loop technique). As described in the reliable transmission / reception system 40 ( FIG. 3 ), the timing information is applied to the encryption device and the decryption device, respectively.

In a system using a cascade or tree analyzer and synthesizer ( FIGS. 1A to 1D), two types of timing signals are required. This is because the inherent delay in the transmission part of the system effectively defines “words” as well as signal bits to be synchronized. Such word synchronization is achieved by providing a second transmitter oscillator. This second oscillator is locked to the first oscillator (instead, the first oscillator is locked to the second oscillator, thus requiring only word synchronization). The second oscillator also operates at a rate that depends on the number of levels in the cascade and the type of system (ie, signal encryption device, signal compressor or modem) provided. Such an oscillator signal may occupy the same bandwidth as the system information. This is because the oscillator signal is subtracted by known techniques for removing signals of known frequency and constant amplitude. In practice, these two types of oscillators preferably provide signals having a special frequency and a special amplitude. In practice, these two types of oscillators have frequencies corresponding to the frequencies that define the upper and lower band edges of the system information frequency bandwidth while maintaining the requirement not to exceed the bandwidth limit of the link. Preferably it is locked to the signal.

  One way to remove the interference signal (ie, the received oscillator signal) from the attached system information signal is that the resulting difference no longer includes a narrowband correlation to the known (received) oscillator. Up to forming a feedback loop that subtracts the amount of known (received) frequency. The operation of the phase locked loop is to multiply the input signal by a local oscillator (which is locked to the received timing signal) and integrate the result with a low pass filter.

Another method of timing synchronization between the receiver transmitter unit of the system is described below in conjunction with FIG.

With reference to FIG. 5 , a system 94 for transmitting digital data over an analog medium includes a modulator-demodulator (modem) 95. Here, only the modulator portion of the modem 95 is shown. The modem 95 has a data assembly unit 96 for forming digital data into frames or bytes having a predetermined length appropriately selected for data transmission. This data assembly unit supplies data to the input port of the synthesizer 98. The synthesizer 98 forms the data into a bitstream in accordance with the sub-synthesizer 20 technique described above with respect to FIG . That is, here, the digital data to be transmitted is applied to the residual inputs W ′ ₁ to W ′ _N of the synthesizer. This data is encrypted by a random number generator (not shown) that provides an encryption sequence. This encryption sequence may or may not be a secret sequence.

  The synthesizer 98 supplies the bit stream to the input port of the DAC 100. The DAC 100 is provided so as to have a non-linear response characteristic, and generates an analog signal corresponding to the bit stream supplied thereto. According to the procedure of the Nyquist sampling theorem, the digitized samples from the transmission side should be converted by the DAC 100 into an analog signal representing the supplied digital signal. This analog signal is then filtered for alias removal by a filter 102 having a low pass filter characteristic. This filter 102 preferably has a relatively steep filter skirt and a cut-off frequency corresponding to half the sampling rate frequency. The filtered signal is coupled to the first input port of summing circuit 104. The timing circuit 106 supplies a timing signal to the second input port of the adder circuit. This adder circuit superimposes two analog signals supplied thereto. Alternatively, the timing signal can be transmitted through the synthesizer 98 input port.

  The superimposed analog signal is transmitted to the receiver 108 through an analog transmission link 107 (eg, a telephone line). Here, the timing signal is used to provide timing data for the receiver so that the bit stream can be recovered from the analog signal. A modem constructed in accordance with the present invention is believed to operate for a transmission link at or near the highest theoretically possible data rate, based on the link's signal-to-noise ratio and Shannon's law.

  In the receiver 108, almost all noise in a frequency band of 1/2 or more of the sampling rate frequency of the receiver is filtered by a filter having a low-pass filter characteristic and a cutoff frequency of 1/2 or more of the sampling rate frequency of the receiver. Should be processed. When transmission link 107 is provided as transmitting a signal having a frequency between 400 Hz and 3200 Hz, the input signal is typically sampled at a sampling rate of about 6400 bps.

  A signal tone, typically having a frequency of about 3200 Hz, is provided by a timing circuit and added to the transmitted signal and further phase synchronized with the receiver as one means of synchronizing the receiver's ADC clock. Since this tone is given to have a known amplitude and frequency, it is subtracted rather than filtered at the receiver. Thus, no data loss occurs.

  Similarly, signal tones, typically having a frequency of about 400 Hz, are used to provide word synchronization to formulate words with inputs to all cascade levels. Signals in the frequency band below 400 Hz can be used for signaling for line turnaround in half-full duplex modems and for transmitting reverse channel data and network information. The bits per cascade level are increased until the maximum power per unit of transmitted data and the Shannon limit are reached.

  In addition, error correction codes such as M in N codes (M redundant bits in N) and scrambled signals are applied to input data words as known to those skilled in the modem art. The

Referring to FIG. 6 , the coded modem 110 uses a direct sequence coding scheme. In this case, each data word modulates all bits in a series of sign bits, and more than one signal shares the link simultaneously. Signal _{S 2} to be shared, for example, audio signals are given as a television (TV) signal or a facsimile (FAX) signal. Alternatively, the signal S ₂ to be shared is given from the added modem of the same type operating in the orthogonal code sequence.

  The coded modem 110 has a coder 112 for providing an encoding operation. The encoded signal is supplied to the synthesizer 114. The synthesizer 114 provides a relatively wideband signal having a noisy frequency spectrum to the input port of the adder circuit 116. The shared signal is supplied to the second input port of the adder circuit. This summing circuit couples the signal supplied thereto to the first end of the transmission line 117.

  In the modulator portion, the signal tap 118 supplies a part of the signal transmitted through the transmission line 117 to the optional signal processor 124. The signal processor 124 supplies the processed signal to the receiver 125.

  The code of the coder 112 is selected to have good auto and cross correlation properties. Thus, modem data is recovered even when the modem operates at low power for shared signals.

  For a receiver 125 provided as a shared device, for example a TV receiver, the modem signal appears as a small background random noise. However, if the shared signal is coupled to the TV receiver 125 through the signal processor 124 containing the code sequence C, most or nearly all of the modem “interference” at the TV receiver 124 will cancel the correlated noise. To be removed by a known description.

The coded modem 110 has a coder 112 for providing an encoding operation C and a decoder 122 for providing a correlation operation C ^. In the correlation operation, the data is recovered by digitally integrating the product of the received sequence and the stored code C. Signals sharing a link typically tend to integrate near zero. This is because the shared transmission is not correlated with the selected code C. An optional interference cancellation operation on the shared signal is performed by the signal processor 124.

A method of timing synchronization between the transmitter and receiver of the coded modem 110 is described below, but it should be noted before describing this method that the method is in the context of the coded modem. Although it can be clearly described within, this method will also be described with reference to other systems such as the signal encryption system described above in connection with FIGS. It can also be applied to signal compression systems.

At least one of the signals W _'k is a sequence having a known correlation properties. Since cascade synthesizers require “word” synchronization, the selected signal W ′ _k preferably corresponds to the signal with the lowest in-band sample rate. However, it should be noted that in a direct sequence coded modem, all W ′ _k are encoded as such.

  There are many examples of suitable code sequences, such as JPL, GOLD code and Walsh code. For purposes of explanation and not limitation, Walsh codes (also known as Hadamard codes) are described. The core of the Walsh code is given as follows.

  Higher order codes are shown below by substituting levels into the kernel.

  This is expressed compactly as follows.

  Many other codes (eg, GOLD codes) are known to have “good” correlation properties. Correlation means that in the case of a binary value with two values (+1, −1), multiplication and integration decrease to the vector dot product of the sequence. The dot product of two identical codes gives a given output (ie (Code1) DOT (Code1) = 4). However, the dot product of two different codes gives a zero output (ie (code1) DOT (code2,3,4) = 0). Similarly, this relationship is also true for each of the other three codes. Thus, these are orthogonal codes.

  Non-orthogonal codes with large autocorrelation and small cross-correlation are also suitable. And some of such codes are known to be particularly good for rapid synchronization acquisition in a sliding correlator. An example of a sliding correlator is made by code3. If word synchronization is unknown (in this discussion, bit synchronization is assumed to be known), one of four possibilities occurs at the receiving correlator. They are as follows.

+ 1 + 1-1-1 Assumed timing of receiver word clock + 1-1-1 + 1 Possibility of received pattern 1
-1-1 + 1 + 1 Possibility 2
-1 + 1 + 1-1 Possibility 3
+ 1 + 1-1-1 Possibility 4

  Correlation, that is, the dot product of the sign of the receiver and each of the four possible patterns, indicates that the correlator calculates the dot products 0, -4, 0, +4 for the four possibilities Show. However, only correct word synchronization (ie possibility 4) has a large positive (ie +4) correlation. By taking the correlation of the assumed clock of the receiver by sliding it bit by bit with respect to the incoming signal, the receiver can find word synchronization. Therefore, the name “sliding correlation” is born. It should also be noted here that maximum correlation occurs when both word synchronization and bit synchronization are correct.

  In order to achieve the synchronization method described above, the residual W ′ of the lowest frequency stage in the transmitting cascade is such that code3 defines its value (or at least the sign of the residual W ′ follows code3). As configured). The previous discussion explains that receiver synchronization can be achieved in some of many ways. In certain system applications, such as encryption, timing from the receiver can also drive other building blocks, such as decryption devices. In full-duplex operation, the receiver clock can also be used for transmission from that end so that there is only one master clock in the system.

Assuming that both bit synchronization and word synchronization are established in the coded modem 110, an incoming data bit is represented as corresponding to either +1 or -1. If the data bit is multiplied by one of the codes, eg, code2, the resulting 4-bit sequence is either code2 or code2 with the sign of each bit inverted. If the sequence is applied as a sequence of bits to one of the cascade or tree synthesizer W ′ inputs, the receiver analyzer will recover that W ′ and correlate with code2. This is to obtain a large positive or large negative number that will determine the output of the receiver as +1 or -1. Here, for simplicity of explanation, an example using a single bit is described. Of course, in an actual system, such operations are typically performed on digital words having multiple bits.

  The application of encoded data bits to the cascade synthesizer's W 'input is somewhat complicated because each stage in the cascade operates at a different sampling rate. Such an operation can be achieved more easily with a tree synthesizer. This is because the input data is assembled into words and applied at once at the lowest stage frequency of the synthesizer. For tree synthesizers, transmitter power is also distributed evenly across the link bandwidth. This is the preferred efficient case. Spreading transmitter energy across the link bandwidth is a recipe for achieving maximum link operation.

Several coded modems of the type described in FIG. 6 can operate simultaneously on the same link while the total link power is constrained. Each modem should use a different orthogonal code. For example, the modem of code3 is prevented from interfering with the modem of code2. It should be noted here that the number of modems sharing a link using the 4-bit Walsh code described above is greater than four. This is because if a group of modems do not operate exactly at the same time, each modem has a different and unique code combination for each of its independent residual inputs. It should also be noted here that these modems are simply given different data bits that are orthogonally encoded.

  A two-wire full-duplex modem is provided by using a set of nearly orthogonal codes. Walsh codes are identified as code1 to code4 above, and their bit position complements are only 1/2 of 16 possible combinations of 4 bits. As shown below, the remaining combinations also form another set of mutually orthogonal vectors. These vectors are numbered c5 to c8 below. The second set of four vectors is not orthogonal to the first set. It can be described as “substantially orthogonal”. This is because the dot product of the members of the first set set1 and the second set set2 is always half the length of the vector. Of course, the dot product of one member and another member of the same group is always zero. However, the product with itself is always equal to its length.

  The master group is orthogonal and the slave group is also orthogonal, but the cross group correlation is -2 for dual and +2 for other cross terms. One end of the transmission link transmits a signal using the master code set, and the second end of the transmission link transmits a signal using the slave code set. It should be noted that the same benefits are realized by using a matrix transformation method that provides synthesizer 114 and analyzer 120.

  Thus, to provide full-duplex operation, the modem at each end of the link can be assigned to use set1 or set2 as master and slave. If the master uses only code 1 and 2 and the slave uses only code 3 and 4, all echo signals are totally canceled by orthogonality, but use the above arrangement based on two sets of codes By doing so, the data for each modem is half the possible rate.

Furthermore, a coded modem of the type described in connection with FIG. 6 coexists with other signals on the link. This is because the correlator provides little or no signal. When the code sequence is lengthened, the data processing capability is lowered, but this effect can be improved. Certain codes other than Walsh codes can make better use of this property for multi-access applications.

The modem 126 shown in FIG. 7 can excite the sub-stages of the synthesizer cascade using a direct sequence code division multiplexing scheme. As described above, such a procedure allows clock recovery based on sliding correlation. It also allows data applied to the code sequence to be recovered at the receiver end using correlation techniques (as is done in direct sequence spread spectrum mode).

  One application of such a coded modem is to take advantage of the processing gain of the correlation receiver to recover the low power signal from the modem buried in a large “jamming” signal. Actual examples of jamming signals include audio (in this case, data is transmitted as “noise” hidden in the audio) and television (in this case, high-definition digital information is standard video for compatibility). Are transmitted on the same channel), code division multiplexing, and 2-wire full-duplex FDX. The disclosed technique is an improvement over these methods. This is because modems use bandwidth more efficiently than ever.

  Further, the modems described herein as being modulators (synthesizers) and demodulators (analyzers) are either baseband (or higher than baseband, except for ADC limitations) RF or sonic transmitter modulators. And may take the form of a receiver. Such receivers have applications for receiving digital high definition TV (HDTV).

A coded modem of the type described above in connection with FIG. 6 and described below in connection with FIG. 7 uses Walsh codes to reserve data applied to the W ′ and V ′ inputs. Encode. It should be pointed out that since the synthesizer itself uses Walsh codes, these two types of coding schemes are independent when implemented with a rotation matrix. For example, the code length of the data input need not be the same as the number of synthesizer outputs. The number of outputs is equal to the length of the synthesizer's rotation operator. Thus, by applying the matrix method to a coded modem, the coded modem is essentially a twice-coded modem.

Referring to FIG. 7 , the coded tree modem 126 includes a plurality of coder circuits 128a to 128h. These coders circuit is coupled to a corresponding one of the plurality of input ports 130a~130h tree-synthesizer 130 to operate in accordance with the principles described in connection with FIG. Output port of the synthesizer, through the link 132 is coupled to an input port of a tree-type analyzer 134 that operate in accordance with the principles described in connection with FIG. A plurality of decoder circuits 136a-136h are coupled to the analyzer output ports 134a-134h to decode the encoded signals supplied thereto.

  In principle, the coded tree modem 126 operates as follows. That is, the coder circuit performs an encoding operation and multiplies the data word by the orthogonal code C. Here, the decoder performs a correlation operation indicated by C ^. It should be noted that the final V ′ input sequence on line 130h is zero if it is assumed that it is lower than the frequency corresponding to the lower frequency limit of the passband frequency of link 312. It is to be set.

Referring to FIG. 8 , a system 138 for transmitting / receiving a compressed signal includes a transmission unit 138a. This transmission unit includes an input signal adjustment circuit 139. Here, the adjustment circuit includes an input filter 140 and an ADC 142. Filter 140 and ADC 142 are selected to operate according to the techniques described above. This is to provide an appropriate stream of digital words to the first input port of the analyzer 144. A quantizer 146 is coupled between the analyzer 144 and the synthesizer 148. In operation, the signal compressor quantizer 146 maps the residual W ′ (N) to a new number that requires fewer bits to describe. This is a compression operation.

  The output signal conditioning circuit 149 has a DAC 150 coupled to the output port of the synthesizer 148. The DAC 150 receives a stream of digital words and provides an analog output signal that represents the bitstream supplied thereto. A filter 152 with appropriately selected filter characteristics couples the analog signal from the DAC 150 to the first end of the transmission line 154.

  The second end of the transmission line 154 is coupled to the receiver 138 b of the system 138. The receiver b has an input signal adjustment circuit 156. This conditioning circuit appropriately filters and converts the analog signal supplied thereto and provides an appropriate stream of digital words to the first input port of the analyzer 158. Inverse quantizer 160 (ie, requantizer) is coupled between analyzer 158 and synthesizer 162.

  In operation, requantizer 160 remaps to the original bit definition. The compression operation, of course, reduces the information content of the signal so that lost information is not recovered. However, in many applications, the lost information is redundant. Alternatively, the level of detail contained in the information discarded by the quantizer is something that a human observer will not feel. Thus, little or no signal degradation is detected.

  The output signal conditioning circuit 163 receives the reconstructed digital stream word from the synthesizer 162 and appropriately filters the analog output signal representing the bit stream supplied thereto into the receiver 138b of the system 138. To the output port.

  In the conversation compression process, the bandwidth is generally reduced by limiting the number of bits for the residual W '.

However, modification to increase Genchijimi bandwidth, first the signal V ₅ on the cascade channel corresponding to the lowest frequency band is replaced to zero, it is achieved by transmitting only the signal W _'5. Next, the signal W ′ ₃ associated with the cascade channel corresponding to one frequency band of the frequency band from 700 Hz to 1400 Hz is erased or coarsely quantized. Furthermore, Huffman coding method or codebook vector quantization method is used in the W _'2.

By adjusting the sample rate, the frequency band of 700 Hz to 1400 Hz is separated. Since human speech, especially English, does not include a format in this band, the cascade channel corresponding to the 700 Hz to 1400 Hz frequency band loses comprehension (by setting the residual W ′ ₃ to zero). It will be erased without giving Similarly, as shown in the table below, W _'1 and W' ₅ is also set to zero.

Thus, in this example, W ′ ₁ -W ′ ₅ and V ₅ are sampled at the rates shown in the table above. Further, for example, only signals corresponding to the residuals W ′ ₂ and W ′ ₄ can be transmitted. These have sample rates of 700 and 2800 baud, possibly less than 2 bits for each residual W ′ ₂ , W ′ ₄ after Hoffman coding. Further reduction is also possible. This is because we can simply consider W ′ ₂ as another sampled signal and thus subdivide by multi-resolution analysis to further reduce the bandwidth.

As an example, 1.5 bits are used for 700 samples / second, and the residual W ′ ₂ is 1.5 bits, 1400 samples / second, 700 samples / second, 350 samples / second, 175 samples / second, and 1.5 bits, respectively. When broken down to 65 samples / second, the total number of bits / second (bps) corresponds to 5085 bps. To this, overhead bits for frame synchronization are added. This method is much less computationally intensive than methods such as linear predictive coding 10 (LPC 10), dynamic excitation LPC and improvements thereof, as known to those skilled in the art of speech compression.

Although not shown here, the receiver is given to have the same form as the received signal applied to V ₀ as well as a clear output taken from V ′ ₀ .

In view of the above, those skilled in the art will now recognize that a combination of the above-described systems is made to form, for example, an encrypted data compression system used on an analog link. This is particularly useful for applications such as transmission and reception of high-definition TV signals where the amount of digital data transmitted exceeds the link's Shannon law limit. Thus, in such an application, the data is first compressed by algorithm means until the sample rate matches the link's Nyquist sampling theorem limit, and then the data is then transmitted to the modem described above, for example with reference to FIGS. Given to one of the systems.

Referring to FIG. 9 , the digital compression circuit 166 includes an analyzer 168 that is coupled to the quantizer 170. The digital signal is supplied to the analyzer input. Analyzer 168 decomposes the signal and quantizer 170 performs the compression operation described in connection with FIG . The quantizer 170 then provides the compressed digital signal to the first end of the digital link 172. The second end of digital link 172 is coupled to the input port of requantizer 174. The requantizer 174 receives the signal supplied thereto, performs an inverse quantization process, and supplies the requantized signal to the synthesizer 176. In the digital method, it is only necessary to perform a logical operation between the W signal and the random number RN, for example, an exclusive OR operation.

Referring to FIG. 10 , the telephone modem 178 includes a synthesizer 180 that can operate in a frequency band of 400 Hz to 3200 Hz. The synthesizer 180 provides a signal to a DAC 182 that typically has a sampling rate of about 6400 samples / second. This DAC converts the bit stream supplied thereto into an analog signal. This analog signal is fed through an analog transmission link 184 to an ADC 186 which also has a sampling rate of about 6400 samples / second. The ADC 186 converts the analog signal supplied thereto into a stream of digital bits. This stream of digital bits is supplied to the analyzer 188.

The synthesizer 180 is provided according to the matrix transformation method described above in connection with FIG . However, it should be noted here, synthesizer 180 is that even given a 3-level tree synthesizer according to Equation Equation3 and 4 described above in connection with FIG.

  In the matrix method, the synthesizer 180 calculates a data frame supplied thereto using a rotation matrix. Here, an 8-dimensional rotation matrix is applied to the data frame. Similarly, the analyzer 188 performs a reverse rotation operation by applying an 8-dimensional matrix corresponding to the inverse of the matrix used by the synthesizer 180.

  As described above, the basic sampling rate of the DAC 182 and ADC 186 is 6400 samples / second. Digital inputs and outputs operate at 1/8 frame rate or 800 frames / second. Each frame consists of, for example, 35 bits divided into 7 words of 5 bits each (or larger) and is applied to channels 180a-180h. The processing capacity is 28,000 bps (35 bits × 800 frames / second). The usable rate is determined by the signal-to-noise ratio on link 184 and forward error correction (FEC). Thus, the modem 178 operates at a data transfer rate (rate) of 28 kbps or less.

  In addition, modem 178 uses FEC. Also, as a general practice, the 300 Hz to 400 Hz frequency band is used for frequency shift keying (FSK) diagnostic signaling.

  Channel 180h corresponds to a signal system of 400hz or less and cannot be used for the data in this example. However, if a signal with constant amplitude and alternating sign is applied to channel 180h, the 400hz tone is filtered out of the line signal by a receiver (not shown) to aid in synchronization. In addition, the known preselected amplitude is recovered as data from channel 188h and used as a gain calibration signal at the receiver, and further used to define block boundaries for the FEC block coding scheme. Sometimes.

Accurate timing synchronization and gain calibration are important to the operation of modem 178. As described above in connection with FIGS. 6 and 7, the coded modem can obtain synchronization information without the tones that appear in the transmitted signal. Thus, the secret communication by encrypting the coded modem is transmitted as low level appearance non-correlated noise within the same narrow bandwidth as the coexisting non-secret communication.

  The data on channels 180a-180g is scrambled as is common practice in modems. As a result, the output becomes more noisy when transmitting ordinary accidental 35 0 or 1 input strings. Without scrambling, a string of zeros produces an output that is modulated but strongly correlated and has no DC component.

  The data received by analyzer 188 is a multiple of the actual data. The magnification corresponds to the dimension of the rotation matrix, which corresponds to the number of channels in the synthesizer 180. The receiver should quantize or round the received channel output to the nearest multiple, and further divide by that multiple to reduce the effects of noise on link 184.

Modems that use baseband modulation techniques in the absence of a carrier signal are given by using the matrix rotation method or by the equations given in connection with FIGS . In either of these approaches, the modem must demodulate and process groups of two or more samples to recover the data. In conventional baseband systems, a single sample is processed to recover the data.

  Thus, in a baseband system, a modem operating according to the matrix method described above modulates and demodulates a group of samples. For example, if the matrix vector length is 8, the demodulator processes the 8 samples together as one independent group. That is, the demodulator multiplies the group of samples by the inverse matrix used in the modulator. It should be noted here that the grouping of samples is part of the modulation and is clearly distinguished from, for example, block data coding methods for error correction, and both techniques can be used with a single modem. It can be used at the same time.

In short, baseband modem technology uses a reversible mapping. According to previous disclosures and the issued parent patent US Pat. No. 5,367,516, modulation and demodulation mapping can be implemented in filter bank synthesizers and analyzers, or rotation and derotation matrices, or baseband. Characterized as the mathematical transformation of and the inverse of it. These characteristics clearly different, but in some instances, describing the same operation on different technologies. The terminology and unity of the three features can be found in the publications cited in connection with the present application. For example, the analyzer and synthesizer of FIG. 1 is a two-channel quadrature mirror filter (QMF) pair described by Vaidyanathan and Hoang. The matrix is the interchangeable polyphase filter matrix of Viterli and Gall, or similar to Bydiana Sun and his students. The inverse polyphase matrix describes the QMF of FIG. 1 and a filter bank with many subbanks, such as a filter bank that is functionally equivalent to the structures of FIGS. 1C and 1D. Finally, the structure of FIG. 1A based on 2-channel QMF is mathematically equivalent to the discrete wavelet transform (DWT), as described by Rioul and Vitterli. The scale function and the residual function are terms (terms) used to describe the wavelet transform.

  Invertable mapping for modems is similar to geometric rotation. In this case, the mapping may or may not have a pure delay due to the filter. The sequence of samples X (0) to X (3) is considered as the coordinates of the vector X in the four-dimensional space. The linear transformation is performed from the X coordinate system to another one coordinate system in the same four-dimensional space. Thus, vector X is converted to vector Y by a “rotation” matrix, by a filter bank, or by a mathematical transformation such as DWT.

  The component of vector X is made equal to a frame of consecutive samples to the transmitter D / A converter. The Y component is assigned to the input to the filter bank corresponding to the subband within the transmission bandwidth. In modems based on the previous disclosure and issued parent patent US Pat. No. 5,367,516, X and Y are two different coordinate representations of the same vector. A modem modulator can be thought of as a linear “rotation” operator [M], and a demodulator as an operator [D]. At the transmitter, X = [M] Y. The vector Y is transmitted so that the received data [D] X = Y when the modem correctly carries the data. This is a condition that matches the necessary condition [D] [M] = z [I]. Here, [I] is an identification matrix, and z represents any pure frequency-independent delay through the system. If there is additional noise at the receiver, the linear demodulation operator must remove that noise term by non-linear thresholding so that it is common to all modems.

  In common sense, including both linear and non-linear operators, all modems need [D] [M] = z [I] to recover the data correctly. However, the aforementioned baseband modem and the issued parent patent US Pat. No. 5,367,516 result in a geometric rotation analogy. This is because linear operators are interchangeable. That is, [D] [M] = [M] [D], or at least they are almost interchangeable over the useful bandwidth of the channel. Two-dimensional geometric rotations are interchangeable, and any dimensional rotation can be made interchangeable by creating a series of two-dimensional rotations that can be interchanged with a series of inverted reverse rotations. This is the essence of the design procedure such as Bydiana Sun for the multi-dimensional (multiband) QMF bank of the cited document.

  As mentioned above and as described in the issued parent patent US Pat. No. 5,367,516, the sampled analog signal can be framed as vector A, and the bandwidth extension is also digitally compressed. It can be digitally encrypted without any need. This is because the encryption device on the transmission side transmits X = [M] [e] [D] A, and the decryption device on the reception side can calculate A = [M] [d] [D] X. is there. In this case, it is assumed that digital encryption [e] and decryption [d] satisfy [d] [e] = [I], and [M] and [D] are interchangeable.

  A passband modem modulates one or more carriers with one or more separate baseband modem waveforms representing data. Carrier modulation and subsequent demodulation is a common linear operation of modern modems. Thus, at a certain carrier frequency, the modem adds sine and cosine carriers. In this case, each carrier is subjected to linear amplitude modulation by a baseband data modulator and a filter. The resulting signal has both a phase and / or amplitude change, as in a quadrature amplitude modulation (QAM) modem. The previously disclosed baseband modulation can be applied in this same manner to form a linear modem for the passband.

  Non-linear modulation to the passband is also possible for data modems and analog signal encryption devices. Non-linear modems are rarely used other than very inefficient frequency shift keying (FSK) modems. However, non-linear modems using the methods described herein are more bandwidth efficient than other linear modulations such as QAM when operating in an area of interest, eg, for wireless communications. In this case, for comparison, it is assumed that there is no forward error correction (FEC) in either the nonlinear design or the linear design. An exemplary non-linear FM modem is described herein.

  The FM DSB signal can be generated by the orthogonal carrier method. In this method, the sine and cosine carriers are amplitude modulated by the integral sine and cosine of the baseband modem signal, respectively. Thus, the amplitude of the cosine function modulates the cosine carrier and the like. In this case, the FM signal is an actual nonlinear modulation and is clearly distinguished from the linear passband modulation. This may or may not include a baseband integrator. FM and nonlinear phase modulation (PM) can be generated by a voltage controlled oscillator (VCO) and by other techniques. As described above and as described in the issued parent patent, US Pat. No. 5,367,516, the data and the modulated signal can be in different coordinates representing the same vector. This perspective provides insights and anomalous techniques for the design of non-linear modems using filter bank and wavelet rotation.

In FIG. 11 , an FM modem 200 according to the present invention is shown in the form of a block diagram. This modem includes a transmission unit 202 and a reception unit 204. Input data to the transmission unit 202 is first divided into data vector expressions by a division element 206. After giving an equivalent average power to the divided signals in general by the nonlinear pre-emphasis amplification 207, the data vector is rotated by the transmitter operator 208 as described above to become a signal vector. After improving the modulation gain by non-linear compression 209 (described below), the signal vector is FM modulated as indicated by block 210 and transmitted by transmitter interface 212 as an output signal to a transmitter (not shown). Given to.

  Typically, an output signal from the transmission unit 202 of one modem 200 is not received by the reception unit 204 of the same modem 200. However, it is possible that the transmission path from which the output signal is transmitted and from which the input signal is received includes a memory device for storing the transmitted signal. In such cases, it is possible for the same modem to perform both transmit and receive functions.

  For example, in a variant embodiment of the invention, the memory device is a magnetic disk or a non-volatile solid-state memory device, and the modulated digital information stored in analog form is later retrieved and demodulated by the same modem. This memory device behaves as a modem transmission link with exceptionally long link delay. As a result, the bit error rate calculation and maximum system transfer rate to or from memory can be determined for any modem system. In particular, the maximum error-free bit rate for data recovery is given by Shannon's law. Here, the signal-to-noise ratio depends on the physical characteristics of the device and the noise figure of the electronic device.

On the receiving unit 204 side of the modem 200, an input signal from an FM receiver (not shown) is received by the receiver interface 214 and sent to the FM demodulator 216. Demodulated signal, after the non-linear inverse compressor attenuator 217 is rotated in reverse by receiving the rotation calculator 218. After complementary deemphasis amplifier 219, the result is ideally assembled by Assembly element 220 to the output data is the same as the original input data. Nonlinear decompressor attenuator 217, in alternative embodiments, further comprising an equalizer.

  A modem using convolutional rotation, discussed above and discussed in the issued parent patent, US Pat. No. 5,367,516, is optimal in terms of potential bandwidth efficiency. Encryption using this technique means that the signal encryption device converts any band-limited signal into the digital domain, digitally encrypts it, and further changes the signal without changing the analog bandwidth. It is optimal in that it can be converted back to the analog domain. This is in contrast to current digital audio encryption devices that rely on digital audio compression algorithms to achieve encryption without bandwidth expansion. Both of these optimal characteristics are the result of the invertible baseband conversion used in the modem described above and in the issued parent patent, US Pat. No. 5,367,516. One preferred embodiment of this invertible baseband transform is a quadrature mirror filter (QMF) bank, also known as a wavelet filter. This allows lossless signal reconstruction across the entire passband of the filter. Since the transition region to the stop band at the band edge can be designed to disappear as the filter delay increases, the modulator satisfies Shannon's criteria for eliminating entropy loss at mathematical limits.

  The rotation operator of the first embodiment is a multi-rate wavelet type filter bank. Such a filter is designed in the same manner as the subband coding scheme. Each analyzer input corresponds to one of the M overlapping subbands. Ports corresponding to the out-of-band region of the signal spectrum are not used for data, but they are used in another embodiment to carry baseband synchronization bits without disturbing the spectrum constraints. . Regardless of the number of subbands, the multiphase rotation matrix is reliably interchangeable across the entire band. There is very little entropy loss at the band edges (not between bands) as determined by the independently chosen filter length.

  For a time frame corresponding to M samples, the input binary information is divided into integer coordinates of the information vector. The output of the demodulation operator is seen as a vector with non-linear coordinates. “Threshold processing” is a non-linear operation that removes noise by rounding received vector coordinates to an integer.

  It is normal modem practice when transmitting data to encode the data into evenly spaced signed odd values. For example, 2 bits are encoded as one of four levels -3, -1, +1, +3. In this case, the threshold processing rounds to the nearest odd number. The modem can also be used to transmit any band limited analog signal. This analog signal may be, for example, but not limited to, a sub-rate including the digital processing steps of digital encryption disclosed above and disclosed in the issued parent patent US Pat. No. 5,367,516. Digitally processed by filtering. When a modem is used in this manner, the modem modulator input samples can be thought of as signed integers that span even and odd values including zero. Receiver thresholding rounds these values. To match the notation, the integer is assumed to be the input to the modulator, the modulator sends the integer to the D / A converter, and the receiver receives the integer from the A / D. The demodulator outputs an integer. This integer is thresholded by appropriately rounding the quotient after dividing by the normalization constant of the rotation matrix to an integer or odd integer. As described above and in the issued parent patent, US Pat. No. 5,367,516, this rotational gain is calculated from the rotation operator (the multiphase matrix of the filter bank).

  Transmission data including the synchronization information is represented by an M-dimensional vector. In one embodiment, modulo addition of the secret vector to the carrier vector in the data representation provides a highly reliable digitally encrypted analog signal from the D / A. The decryption modulo subtracts the secret vector to recover the transport vector.

  Input information bits are assigned to the data representation coordinates of the vector, and each coordinate is multiplied by a unique pseudo-random sequence at the chip rate. This chip rate result is input to the modem's in-band data port where it is up-sampled by a factor equal to the number of dimensions of rotation. Modem conversion results in the orthogonal addition of each up-sample filtered subsequence to the overlapping subbands of the transmitter spectrum. Instead, a single diffusion function is applied to the information. This information is divided according to the desired modulation efficiency and becomes a component of the vector. This vector is transformed by the wavelet filter bank.

  Cooperating receivers filter the input to the subband by derotating the signal vector back to data space coordinates. However, the effect of the diffusion function must be removed prior to thresholding. This is because the spread signal has a signal-to-noise ratio less than one. “Despreading” is accomplished by correlating a known spreading function with unthresholded data representation coordinates. The vector in the transmitter data representation is redirected anywhere in the in-band signal space by an invertible secret transform (ie, digitally encrypted by the modulo addition of the secret vector), so it is truly high A reliable spread spectrum is made. The receiver must decipher the unthresholded data representation prior to despreading. The receiver performs reverse rotation, decoding, despreading and thresholding in this order. Geometrically, the encrypted vector is always within a radius R, which is the radius of the N-dimensional signal space. Modulo R vector subtraction preserves a large non-integer noise vector added to the encrypted carrier vector so that it can later be correlated (despread) and thresholded.

Thus, again, the N-dimensional space is defined by N samples to the D / A converter. The data coordinate system is selected such that the subset n in the N coordinate of the data representation defines all of the information points in the band. For each of the n coordinates, B-bit data is transmitted as one of 2 ^B integer levels with an efficiency of 2 * B bits / Hz per coordinate (ie per subband). Since the subbands overlap completely within the passband and typically have a -70 db stopband externally, the overall bandwidth efficiency is compared to the total bandwidth where the filter transition region is used for data. 2 * B bits / Hz for nominal purposes. The remaining N-n data coordinates are not used for data, but can be used to synchronize modems.

  Thus, a data vector having integer coordinates is rotated to signal coordinates by a first exchangeable operation. From a geometric point of view, this operation is to apply a multiphase matrix for data vector mapping (or rotation) to the signal vector representation. Alternatively, the mapping is performed by using a finite impulse response (FIR) filter bank or an infinite impulse response (IIP) synthesizer. Instead, the mapping may in some cases be a wavelet transform that is shown to be mathematically equivalent to the filtering performed by the synthesizer. The first exchangeable operation at the transmitter is reimbursed by the second exchangeable operation at the receiver. Together, these operators produce an identification matrix that allows complete recovery of the input data. The actual FIR lattice filter implementation of a fully reconstructed (PR) QMF bank is the identification matrix multiplied by a scalar gain and a pure delay. Since the matrix is typically an unnormalized integer format, a gain factor occurs, and a pure, frequency independent delay factor represents the delay through the FIR filter. Near perfect reconstruction (NPR) filter banks are also known to those skilled in the art of QMF design. The FIR grating QMF is designed by using the interchangeable properties of geometric rotation. The FIR transverse filter configuration is determined from the lattice configuration, but in doing so, the PR characteristics are leaked and the filter bank gives NPR. However, by using computer-aided design techniques, NPR filters are usually optimized to provide better overall characteristics for stopband attenuation than, for example, PR grating filters used to initiate iterative design. The NPR filter is provided. Thus, the mapping used to construct the baseband modem modulator and demodulator of the above disclosure and parent patent US Pat. No. 5,367,516 is completely Or it is interchangeable in terms of near-complete reconstruction.

The two-dimensional QMF as shown in FIG. 1 is described in literature by a multiphase matrix. For example, if the high pass and low pass filters are FIR, each filter transfer function is factored into even and odd powers of the Z transform variable. Therefore, if H (z) describes a high pass filter and L (z) describes a low pass filter, the filter of the QMF analyzer is factored as follows.

  This can be written in the following matrix form:

The sample rate change in the QMF is combined with the delay matrix on the right side to become a serial / parallel conversion for the analyzer. The 2 × 2 matrix on the right side, whose elements are shown as column vectors, is a multiphase rotation matrix. This sample rate change also converts a power of z ^{2 to} a power of z, but the z variable is often omitted because only the coefficient of the power of z in the column vector is related to the computed value. A similar definition is used for the QMF synthesizer. Up-sampling at the input to the two-band synthesizer means that only odd coefficients contribute to odd output samples. Thus, the filtering process proceeds at a rate of 1/2 using a 1/2 length filter. This is what the multi-phase notation describes. A multiphase matrix for a filter having more than two subbands is defined in a similar manner in the cited reference relating to the parent patent US Pat. No. 5,367,516.

The multiphase matrix is further factored into a form corresponding to a FIR cross filter. This filter works on vectors, not scalars. That is, multiplying the square matrix C _J is power each having a scalar element multiplication j of z. Thus, the modulation operator [M] is described as follows.

In this form, the modem frames the data bits and converts them to vector coordinates. This vector is input to the vector filter and rotated. The previous L-1 data vectors are stored in the delay line of the transverse filter. The vector filter uses the matrix C _j to map the current and L−1 previous data vectors to the new vector, and then finds the resulting baseband modem output vector by vector addition. The receiver uses a vector filter corresponding to [D] to recover data by reverse rotation. In one embodiment, the hardware ASIC uses a single time-division 9-tap filter with a coefficient ROM bank and a bank of shift registers to perform modem rotation.

  This vector filter arrangement is of course mathematically equivalent to other forms of rotation. However, the transverse configuration uses the same method that the modem described in the above discussion and parent patent US Pat. No. 5,367,516 applies to known modem field transverse equalizers. This suggests that it can be equalized against transmission link distortion. The equalizer filter and the demodulation filter [D] can be significantly the same.

  Thus, different embodiments of the present invention use an adaptive adjustment of the receiver multi-phase matrix, also called the derotation matrix. If it is determined that some form of frequency dependent distortion has been introduced into the transmission path, the method of the present FM modem, or as described above or found in the parent patent US Pat. No. 5,367,516, The polyphase filter in the receiver portion of any modem used can be adjusted to compensate for such distortion. Another way to describe this is that some of the inverse rotation coefficients in the inverse rotation matrix are adjusted to compensate for frequency dependent distortion in the transmission path. This adjustment minimizes the calculated error function. This is the same as the design procedure for optimizing the NPR in the design of the filter bank. However, for modems, the NPR solution does not accurately reconstruct the information. This is due to amplitude distortion and delay distortion of the modem transmission link. The receiver multiphase filter bank in the FIR implementation is equivalent to each subband FIR filter. Here, the filter output is decimated and thresholded to recover the data. This is known to those skilled in the field of modem equalization but is in the same form as a fractionally spaced equalizer. Therefore, methods in the field of modems, such as, but not limited to, a least mean square (LMS) algorithm for adaptively adjusting equalizer coefficients in proportion to the error function, are included in the demodulator. Can be applied to multiple phase matrix coefficients. This is to provide the receiver with an adaptive reverse rotation that corrects for link distortion without using an independent equalizer filter. Thus, a piecewise sampled FIR filter is provided. The receiver's multi-phase filter adjustment is made at the beginning of the communication and is left at a certain setting. Alternatively, as an example, it is dynamically adjusted by monitoring the RMS spread of noise for integer or odd thresholds from the demodulator.

Additional noise on the transmission link between the transmitter and receiver adds a noise vector to the transmitted signal. As a result, the reverse rotation made at the receiver results in a recovered data vector having non-integer coordinates. In the first simplified embodiment of the present invention, the threshold operator employs the nearest integer coordinate value as the most probable symbol . In more complex receiver embodiments, thresholding is performed by a Viterbi algorithm. This is because the rotation is a convolution with Mn extra degrees of freedom. This allows “free” error correction without the use of parity bits.

  In each transmit frame of a modem according to the above disclosure and parent patent US Pat. No. 5,367,516, each D / A sample depends on a scale input to the modulator and all other residual inputs. In order to avoid aliasing caused by conversion to and from analog by D / A and A / D, the preferred modem design does not use the highest frequency subband to transmit data. Since the number of samples to / from the converter is equal to the total number M of subbands, the transmitted signal is redundant. Other subbands may optionally be omitted to finalize the output spectrum, for example to avoid sending DC. This further increases the redundancy to n / M samples. With the technique of alternately changing the sign of the fixed amplitude, for example, a synchronization signal transmitted in the highest frequency subband does not carry information and therefore does not increase redundancy. If the receiver thresholds a series of frames without bit errors prior to the current frame, the receiver uses those results to help demodulate the current frame. For example, an unthreshold demodulator output is between two allowed odd thresholds. In order to determine which soft decision value is most probable, the receiver uses the preceding data frame along with two possible current frame soft decisions to generate a modulated signal at the receiver; It is possible to correlate with the actual received signal. The receiver then makes a final decision based on a better correlation as to which integer level was most likely transmitted for the current frame. This procedure, implemented through dynamic programming with the Viterbi algorithm, is possible because of the redundancy inherent in the convolution output of the above disclosure and the parent US Pat. No. 5,367,516 modulator. This is substantially different from the soft forward error correction (FEC) method known in the modem field as Trellis Coded Modulation (TCM).

  In TCM, redundancy is provided by attaching one or more suitably calculated parity bits to the data prior to transmission, as is commonly practiced. The coding gain according to the trellis coding method must make up for the energy loss per bit due to the transmission of surplus parity. Furthermore, the TCM modem cannot be optimized. This is because Shannon capacity is sacrificed to transmit parity bits. This does not exclude the use of TCM according to the invention. Other methods of forward error correction known in the modem field, such as multidimensional coding methods, are also applicable to unique multidimensional signals generated by coordinate rotation. Thus, a rotation-based data modem that normally uses only odd coordinates for the data vector would instead use a technique that reduces the probability of bit errors from a large set of even and odd pairs of successive vector coordinates. You can choose.

  What is implied in the definitions of both the modem and the encryption device is the assumption that the data is rotated into an N-band analog signal. N samples of D / A are within a precise time frame. Placing strict simultaneous requirements on bandwidth must be done with care by known conversion techniques. This is because sine and cosine based transformations are complicated by the infinite magnitude of their waveforms. When time is definitely constrained, the frequency component extends indefinitely. The reverse is also true.

  Wavelet theory gives a transformation based on a “mother wavelet” that has no infinite magnitude in time. Similar to Fourier analysis, many basis functions are summed together to represent an arbitrary signal. The mother wavelet is extended and shifted in time to form a set of residual basis functions and an associated set of scaling basis functions. Thus, a single mother wavelet produces one transform, just as a prototype sine wave produces a Fourier transform. There are countless “mother” wavelets, each producing a different transformation. The wavelet transform is quite real (no imaginary component or carrier). This halves the complexity of modulation and equalization. Since only the real component is involved, it is possible to frequency modulate the input data stream without having to consider the virtual component.

  An exemplary FM modem according to the present disclosure uses an 8-dimensional multiphase filter as a baseband input to the FM modulator. As shown in the table below, a non-linear distribution of bits per band is used.

Thus, in the presently shown embodiment, there are a total of 13 bits per input symbol . One design goal is that each band has approximately the same power (where RMS average = 9 db). According to the following equation, the noise coming from the receiver appears to depend on the square of the subband frequency. The power spectral density PSD is given by Couch's “Digital and Analog Communication Systems” 4th edition, Macmillan, equations 7-125.

Here, K is the gain of the FM detector, A is the carrier amplitude, and N ₀ is the double-side noise power spectral density.

  According to this frequency-to-noise relationship, fewer bits are transmitted within each subband as the subband moves away from DC. This is because noise increases with the square of the frequency. Therefore, it is possible to transmit a large number of bits at the lowest frequency. This is because the noise coming from the discriminator in the receiver is significantly less than in the high subband. In the above example, subband 0 has 4 bits, while subband 6 has 1 bit. Also, the number of levels in each band is selected to fit the FM discriminator parabolic noise density function PSD.

  In the first embodiment of the present invention, it is possible to provide the FM modulator with a subband having a decreasing average power level and an increasing subband frequency. For example, the highest subband can carry fewer bits, but all subbands use the same level of spacing per bit. In the second embodiment of the present invention, the levels representing each bit or bits in the subband are separated or pre-emphasized. This is to provide an equivalent average power across all subbands. In other words, voltage levels of +/− 1, 3, 5, 7, 9, 11, 13, 15 are used in subband 0. This is to represent the data carried by the 4 bits assigned to this subband. A voltage level of +/− 9 is used to represent a single bit state 0 or 1 assigned to subband 6. The average power for these two subbands is thus approximately the same. This technique of placing levels apart in high frequency subbands is referred to as a pre-emphasis technique. A signal pre-emphasized in the data coordinate representation by this method is converted into the signal coordinate representation by a multi-phase coordinate rotation filter. This sampled analog signal can be used as an input to an analog FM transmitter. The receiver detects these levels and converts them back into bits. Thus, no explicit pre-emphasis filter or de-emphasis filter is required.

In the above example, the level for each subband is given as an integer (ie +/− 1, 3, 5...). In other words, each subband has 2 ^B levels. In different embodiments of the application, the number of levels in a given subband is not a power of 2, but if the total K bits are transmitted in the p subband, the total number of levels in the p subband is 2. ^K. A binary mapping algorithm is used to determine which of the 2-n levels are represented. By adjusting the total number of bits per subband and bit per symbol , the bandwidth can be optimized for the carrier-to-noise ratio. Computer aided design iteration provides these optimization values.

  In order to compensate for the high frequency sub-band pre-emphasis, it is necessary to “de-emphasize” the high frequency signal received at the receiver. Although much data is not transmitted in the high frequency subband, the data is carried at a high level. This is to provide uniform power from the transmitter to the entire subband. Attenuating high frequency subbands results in de-emphasis gain Gd. This de-emphasis produces a total gain approximated by the following equation for the M dimension.

Here, G _S is either a gain of carefully designed modem vary slightly for each subband in the design having a level which is limited to 2 ^B levels in each sub-band. Obviously, multiple subbands M increase the gain. Further gain improvements are possible by forward error correction and by reducing the peak-to-average power ratio PAR using decompression or controlled vector filtering. With the latter technique, the output peak voltage is pre-calculated at the transmitter, and surplus bits that do not carry useful information are transmitted with the data. These surplus bits are selected in a manner that reduces the peak-to-average power ratio (PAR) from the baseband modulator.

  Assuming that H (i) is pre-emphasis amplification in subband i, subband i = 0 to M−1 is as follows.

Here, the denominator in the above formula for G _S is proportional to the integral of the sub-band i for noise PSD from the discriminator to subband i + 1. Subband i = 0 carries k-bit data and does not carry pre-emphasis. Therefore, H (0) = 1. Subband 1 carries k-1 bits with H (1) = 2. The same applies hereinafter. A typical assignment for M = 8 yields approximately equal power in each subband.

Total B = 26 bits per symbol . Various bit level assignments are shown in the above example of bits per subband.

  The total gain G includes the de-emphasis gain Gd given above, the FM transmitter gain Gm (also known as the modulator gain) given as follows due to the FM index, and the noise reduction gain Gr. It is demanded from. The total gain G is given as follows.

  Here, G ′ is equal to the minimum de-emphasis gain Gd from the de-emphasis gain equation given as described above. An exemplary modulation gain factor Gm is given as follows.

  Where PAR is the peak to RMS voltage ratio to the frequency modulator and m is the modulation index.

FM index is the ratio between the peak frequency of the peak carrier frequency polarization shift and the base-band signal. As shown in the previous equation, as the peak-to-average ratio (PAR) increases, the FM gain, or modulation gain, decreases. To address this issue, another embodiment of the present invention uses non-linear amplification of the baseband signal prior to insertion into the FM transmitter and complementary inverse amplification at the receiver. This is collectively referred to as stretching. In a preferred embodiment, the logarithmic Mu modulo function digitally amplifies the transmitter by table lookup. The Mu method is applied to signal coordinate representation. In a preferred embodiment for a large carrier-to-noise ratio (CNR), the data bits are framed, represented in data coordinates, then pre-emphasized, rotated, Mu-amplified by lookup, and as a sampled sequence. Applies to FM modulator. As described above, in one embodiment, the inverse compression attenuation includes an equalizer.

  In another preferred embodiment for a small CNR, the data is framed and represented in data coordinates, then rotated to signal coordinates, digitally Mu amplified, back rotated to data coordinates, pre-emphasized, Rotated to signal coordinates and applied to FM modulator. This latter configuration allows de-emphasis immediately after the discriminator and is preferred when operating near the FM threshold or when operating in the presence of large amplitude non-Gaussian interference. A preferred Mu method function for non-linear amplification of the transmitter is expressed as:

  Here, the input Vi has a maximum voltage value Vc, and the output Vo has a maximum value Vp. The value of Mu is greater than or equal to 1. Typically 255. The FM excursion is determined by Vp, although Vc reflects the accuracy of the rotation filter calculation.

  A second advantage of logarithmic amplification is the result of the logarithmic inverse amplification (ie, decompression) required at the receiver, which is equivalent to attenuation. The input signal is amplified logarithmically at either end of the transmission link and then de-amplified. However, noise has been introduced into this transmission link. Therefore, this noise is attenuated anti-logarithmically, resulting in a noise reduction gain Gr, as shown in the receiver gain equation above.

  The bit error rate of a modem depends on the energy per bit. In other words, this energy separates each data representation level. This rate also depends on the noise energy density. The signal-to-noise ratio of the FM modem is as follows.

  Thus, unlike a linear modem, the efficiency is continuously dependent on the design value for C / N. The efficiency of a linear modem changes step by step, for example when applying a modulation level from 4-PSK to 8-PSK. For a given design, increasing C / N improves the bit error rate (BER) for both linear and nonlinear modems. However, a small increase in FM modem C / N can instead be used to increase bandwidth efficiency or to decrease bandwidth with the same efficiency. That is, the increased C / N can be used to decrease the FM modulation index. The FM modulation index reduces the bandwidth according to the Carson rule. This rule specifies the FM or nonlinear PM bandwidth W as follows.

  Here, w is a baseband bandwidth. Thus, non-linear modem systems can reduce adjacent channel interference (ACI) while maintaining a constant BER and efficiency by using a C / N excess relative to the design C / N value. Alternatively, the modem can increase efficiency with the same BER and ACI.

  The modem signal-to-noise ratio is Eb / N0 multiplied by the modem bandwidth efficiency (given in bits / second / Hz), as follows:

  Here, Eb / N0 reflects energy per bit versus noise energy density.

The application of FM transmission techniques to the aforementioned data transmission benefits from the fact that amplitude distortion is irrelevant to FM. This is because information is retrieved from the frequency, for example, by the receiver tracking the zero cross point. In FIG. 12 , data transmission by FM double sideband is depicted in the form of a flow diagram. In step 230, the data to be transmitted is multiphase filtered as described above. Also, as described above, FIR filtering includes the addition of a given number of previous multiphase filtering processes. This point will be described in connection with FIG. 14.

As is generally recognized, frequency modulation is achieved by 1) directly adjusting the frequency of the carrier, or 2) adjusting the phase of the carrier with a signal that is an integral of the information to be transmitted. The The use of phase adjustment to achieve frequency modulation is preferred in one embodiment of the present invention. This is to make available a chip that allows phase modulation in a completely digital manner, thus maximizing the reproducibility of the input signal. The phase adjustment FM is shown in FIG. 12 for the FM double sideband and in FIG. 13 for the FM single sideband.

  In FIG. 12, the multiphase filtered input is integrated at step 232. In step 236, a lookup table is used to digitally simulate the sine carrier signal. This carrier signal is then phase modulated using the result of the multiphase filter integration (step 238). Since the zero cross point of this signal contains encoded information, counting the zero cross point (digitally-step 240) provides the desired information to the RF switching amplifier (steps 242 and 244). . In the first embodiment of the present invention, the switching amplifier is a class C amplifier. In the second embodiment, a VCO is used to generate FM in the passband. In the third embodiment, FM is generated using in-phase and quadrature techniques.

  The FM double sideband offers significant advantages over prior art modems. It provides immunity to noise associated with amplitude modulated signals. Distortion due to nonlinear amplification is avoided. The RF amplifier used need not be precise (and expensive). This is because the amplitude modulation need not be accurately reproduced. It also gives some degree of circuit simplicity. This is because carrier recovery is not required at the receiver.

However, the FM double sideband provides half bandwidth efficiency compared to the FM single sideband. Therefore, another embodiment of the present invention uses FM single sidebands. The flow diagram of FIG. 13 shows the FM single sideband for the present invention. In step 250, the input data is multiphase filtered using, for example, an FIR filter or wavelet transform as described above. This step is further developed in FIG. 14 described below. The polyphase filter output is then integrated and the result stored (step 252).

  The Hilbert transform processes the input signal and effectively phase shifts the input by 90 degrees (step 254). The output of the Hilbert filter process is similarly integrated (step 256). The Hilbert filter integration is then entered into an exponential lookup table (step 258).

  The sine carrier signal is digitally simulated using a look-up table at step 260. This carrier is then phase modulated using the integrated polyphase filter output (step 262).

  According to the FM double sideband scenario, the zero cross point of the phase modulated carrier is calculated (step 264) and a digital pulse train corresponding to the zero cross point is generated (step 266). Finally, in the case of a single sideband, a switching amplifier whose input current is proportional to the output from the exponential look-up table amplifies the digital zero cross pulse train (step 268).

For either FM double or single sideband, the first step of the transmission procedure, ie multi-phase filtering of the input data, is accomplished as follows. Referring to FIG. 14 , the input data is first divided into blocks having B bits, where the B bits are further separated into M coordinates of the vector (step 270). In the illustrated flow diagram, B = 13 and B = 8, but other values for these variables can be used. Since the de-emphasis gain factor varies with (M ³ / (M−1)), an increase in M allows a decrease in the carrier to noise ratio. The optimal number is determined empirically and always affects the determination of pre-emphasis for the entire subband. This is as discussed above in relation to the de-emphasis gain factor.

The stack is used in step 272 to hold the current vector and the previous L vectors. As shown, the first embodiment uses L = 9 vectors. These L vectors are multiplied by an M × M dimensional scalar matrix to realize the initial multiphase filter processing (step 274). As shown, this scalar matrix is 8 × 8. Subsequently, in step 276, matrix multiplication is performed on the previous (L−1) vectors stored in the stack using a different 8 × 8 matrix for each iteration, and the result is accumulated. Is output (step 278). As described in step 280, the previous steps shown in FIG. 14 are then repeated at the symbol rate (R / B) to transmit R bits / second.

  In all of the above description, it is understood that an alternative procedure for FM modulation of the input data can also be performed with this multi-phase technique. However, the described procedure maximizes the percentage of data operations that are digital.

Referring to FIG. 15 , a procedure for receiving FM transmission data is shown. In step 284, the received signal is low noise amplified and passed through an image filter. This is to erase the image introduced by mixing the received signal with the local oscillator. Next, in step 286, the filtered received signal is downconverted to an intermediate frequency and in-band filtered. The zero crossing point of IF is counted in step 288. There is no need to use FM modulation to recover the carrier signal or determine the phase of the carrier.

  The DC filtering of step 290 is useful for the application of the present invention and is subject to substantial Doppler shift. For modems used in low earth orbit satellites where the Doppler shift follows known orbit parameters, filtering is used in one example to eliminate such DC or low frequency distortion. Another way to deal with such low frequency distortion is to avoid using the lowest frequency subband. The same step of decimation filtering facilitates the counting of zero cross points. The latter needs to be at a high rate for sufficient signal resolution, but it introduces unwanted noise. In step 292, polyphase filtering involves the application of the interchangeable operator used in the first step of the transmission procedure.

  Threshold processing is used to discriminate between reception levels. The transmission signal is transmitted at one of a plurality of easily distinguishable levels. However, noise introduced in the transmission link shifts the received signal to an intermediate point in the expected level. Therefore, in step 294, it is necessary to assign the decimated received signal to one of a plurality of signal levels.

  Although carrier recovery using FM data transmission is not required, baseband synchronization requires some form of bit synchronization recovery (step 294). As mentioned earlier, such synchronization information can be transmitted in subbands that are not used for data transmission. If Doppler shift is predicted to be a problem, for example, in low earth orbit satellite communications, the lowest frequency subband is not used for data but is available for synchronous transmission.

  Similarly, the highest frequency subband is typically not usable. There is no filter at all sufficient to avoid aliasing in equivalent A / D and D / A conversion operations. By transmitting a known bit pattern at the 3db point between the second highest subband and the highest subband, the synchronization information is within the bandwidth of the transmitted signal, but not the available data bandwidth. . Such synchronization information can be inserted by sampling the DC signal at the synchronization bit rate. High-pass and low-pass filtering produces a sine wave output at that sample rate. Each square wave is at the high or low end of the respective subband. Thus, this synchronization signal is orthogonal to all of the data signals, similar to the pilot tone.

The specific steps involved in the application of the operator in step 292 are shown in FIG. Specifically, the sampled input is divided into vectors having the same number M of dimensions as used in the transmission procedure (step 300). In the illustrated example, M = 8. This result is pushed onto a stack of L vectors (step 302). In this example, L = 9 is used. The first vector on this stack is multiplied at step 304 with the same 8 × 8 scalar matrix used in the transmission sequence. This is repeated for L-1 previous vectors on the stack and the results are accumulated (step 306). This accumulated result gives the coordinates of the result vector total (step 308). This results in the original transmitted data after thresholding and recovery in step 294. The steps of FIG. 16 are repeated at symbol rate R / B to receive R bits / second (step 310).

  The steps described above show that this is an all-digital FM modem for all practical purposes. The non-digital portion modulates the current input to the class C switching amplifier using the passive tank circuit used in steps 244 and 268, the receiver front end of steps 284 and 286, and an exponential look-up table. It is only the D / A converter necessary for this. The transmitter RF power amplifier is actually a digital switch. The filter can also be implemented so that all multiplications are binary shifts and additions. Thus, ASIC costs and power loss can be reduced.

  The previous description and equations generally assume that only one of the M subbands is not used to prevent aliasing in the analog converter. In an alternative embodiment, the speed advantage arises because more than one subband is unused.

  As previously recognized, this technique and general structure can be applied to transmission of multi-phase filtered data using AM modulation in different embodiments. The adaptive rotation of the reception rotation matrix can be applied to an AM modulation modem using the LMS algorithm as described above. For the LMS method, the column of the vector filter tap matrix is iteratively corrected by subtracting the vector proportional to the error multiplied by the input vector stored in the filter delay line for that tap. The error is either the difference between the unquantized output of the modem receiver and the quantized output (ie, error margin) or the difference between the unquantized output and the known training data sequence. These general techniques are similar to those used in modem equalizers. The exception is that they are applied to the rotation matrix without an independent equalizer.

As described above with respect to one embodiment of the modem disclosed herein, the parallel stream of input data is a vector that has been rotated using a multirate filter bank prior to transmission over the data link. Under ideal conditions, the data link introduces no distortion, the impulse response of the filter is a matched wavelet, and the receiver multirate filter bank behaves as a counter-rotation in vector space. Ideally, sampling at the symbol rate of each band of the filter bank of the receiver effectively calculate the autocorrelation of the incoming signal. To completely recover the original data by achieving wavelet transform, like all other bands, all early and stored in the filter from the past and future symbols in the same band or cross correlation of the received signal for later duplicate wavelet is zero, least must also be a "self-interference amount" negligible. In other words, the cross terms must be orthogonal. Nevertheless, the data link distortion disturb the orthogonal, also cause the self-interference.

  The transmitter transmits frequency reference “pilot” tones to the receiver, eg, in subbands that are not used for data transport. At the receiver, a phase locked loop (PLL) can be used. This is to synchronize the phase and frequency of the received signal with the correlator of the receiver so that only a small residual phase error exists in the PLL band. Links that include analog conversion filters for the D / A and A / D converters in the transmitter and receiver, respectively, do not have a linear response that varies (ie, varies) in phase from band to band.

To accommodate such offset, the cross-correlation for the set of wavelets are computed by the filter process. The transmitter and receiver multi-rate filter banks in one embodiment have filter coefficients that are wavelet functions W (I, J), as described above. If there are M subbands in the filter, I is in the range 1 to M, J is the time of the current symbol at the transmitter, and j is the time when the symbol reaches the receiver. For digital processing, time is measured in samples, so I and J are integers, and i is a special subband processed at the receiver.

The modem transmits symbols separated by time T for each of the M samples. As a result, symbols are transmitted as... J-3T, J-2T, J-T, J, J + T, J + 2T. The symbol transmitted at J arrives at the receiver at time j. Thus, singular symbols reach j + nT for processing at the receiver. A wavelet has a limited period and spans only N symbols . As a result, the “n index” is limited to a small range between − (N−1) and + (N−1). This is to cover all wavelets that overlap in time with the received symbols at time j.

  The frequency of the subband is well defined so that any subband i overlaps with only the adjacent subband. Thus, it is sufficient to consider only the band i + m. Here, m is in the range of, for example, -2, -1, 0, +1, +2, and there is no portion where the frequency overlaps over the two adjacent subbands closest to each side. That is, for a magnitude of m> 2, this filter is in its stopband. Of course, i + m is always in the range of 1 to M, where M is the total number of subbands in the filter.

The transmitter encodes the bits in the data symbol D (I, J) and transmits an amplitude weighted wavelet for each band. The symbols are transmitted as follows:

  The sum of these terms for all active subbands reaches the receiver at time j. Here, j = (J + delay in link). Considering all subbands except m is not significant for receiving band i.

  Thus, when the sum is taken for all allowed m and n, there is a series of wavelets with overlapping time and overlapping frequencies.

The receiver adjusts its timing and, if necessary, the phase of the wavelet of the receiver filter. As a result, the receiver filter evaluates the inner product with W (i, j) for each symbol period T to obtain:

The notation <,> represents the inner product. The constant C ₀₀ represents the autocorrelation of the wavelet, that is, the inner product with itself. As a result, data symbols _can be found by dividing the _{C 00.} Thus, it becomes as follows.

Here, D (i, j) is a transmission symbol D (I, J) delayed in time. The wavelet functions are reliably or almost orthogonal. Therefore, in the receiver, the cross-term inner product, by design, or all zeros, very small compared to C _00. Receiver precisely recover the symbols from the normalized recovered symbols (Normalized Recovered Symbol) by threshold detection.

  All relevant wavelet correlations or dot products can be written for band i as follows:

Subscripts represent adjacent bands and adjacent symbol times, as defined above, with positive and negative integers in the range.

However, dispersive impairments within the link create a relative phase shift between subbands. In that case, the “orthogonality” of the wavelet is lost and a cross term appears as a self- interference in the recovered symbol .

In the disclosed invention, a fixed tap weight filter bank is used to approximate the cross term to a special differential (differential) phase shift. The output of the fixed tap weight filter bank is weighted by a known function F (p) of the measured differential phase shift, and the weighted filter output is subtracted from the recovered signal to remove self-interference. . In the preferred design, the fixed coefficients of the fixed tap weight filter bank are calculated from the dot product at one value of the differential phase shift. If the phase cannot be measured, an adaptive combiner can be used.

In the preferred embodiment, the special differential phase shift is 90 degrees (90 degrees is one predetermined value) and the known weighting function is sin (p). Here, p is the measured differential phase angle.

If the phase of band i + m is angle Pm, the adjacent band has a differential phase p = P ₀ -Pm. The inner product for the differential phase shift p is first calculated as follows.

The band i self-interference has several components. For example, the signal recovered in band i is input to the filter S. i left band, namely band i-1 recovered signal from is inserted to the band filter A _L next, and the signal recovered from the i + 1 is inserted into the right side of the right filter A _R of the band i . The weight function F (p) is determined from the p dependence of the X coefficient. Self-interference is eliminated by combining with the delayed recovery signal R as follows.

  The principle of calculating the self-interference term from the wavelet inner product can be extended from a phase shift fault to include a time shift fault. In this case, the time shift t is calculated as follows.

  Self-interference due to timing offset is removed by combining the filtered signals in the same way as for the phase, but the filter tap weights are selected from the Y values.

  Even though the sliding correlation can remove the timing offset between transmitter and receiver for one subband, there is also an offset in the other bands due to group delay changes (ie, variance).

  The foregoing describes a method for providing phase and time equalization for a wavelet modem based on the cross-correlation of wavelet functions. As mentioned above, an adaptive combiner can be used in examples where phase or time shift cannot be measured easily. Furthermore, since one m-band modem has only m phase shifts relative to the correlator reference phase, all of the m bands are recovered with only their m parameters. If the m angle can be measured, the combiner need not be adaptive. Nevertheless, distortion or frequency offset within a wide subband complicates the cross-correlation calculation, in which case all taps are required to be adaptive.

In other applications of the dot product cross-term, an isolated wavelet W (i, j) that does not overlap in time or frequency with other wavelets is offset in time by u samples from the receiver. To reach. The receiver calculates the isolated symbols as well as small early and late “satellite” responses in the symbol time before and after the true symbol .

The receiver can perform a sliding correlation as disclosed to find the correct time. However, the accuracy u of the wide correlation peak of the isolated signal for small time errors does not allow a reliable determination of the setting where u = 0. However, if the correlation is made, for example, one symbol time before and one symbol time after the true arrival of the peak, there are two more inner products.

  If the wavelet is symmetric or asymmetric, it is desirable for the modem's linear phase response, but if the timing is correct and the phase of the wavelet of the inner product matches, the early and late signals Amplitude is equal. That is, when u = 0, the satellite response from the receiver filter has the same magnitude when synchronized. Furthermore, the magnitude of the early and late signals relative to the isolated signal can be used as a threshold to ensure that the sliding correlation is close to u = 0 and not close to the false peak. .

  If the phase of the received signal relative to the receiver wavelet is unknown, the receiver also calculates the cross correlations Early 'and Late' from the wavelet w (i, j). Here, w (i, j) is obtained by shifting the phase of W (i, j) by 90 degrees. In this case, the energy, specifically the energy squared, is:

The Early and Late energy or energy square is equal when u = 0, ie when the transmitter and receiver are time synchronized. Thus, the incoming isolated wavelet aligns with the receiver wavelet. The transmitted symbols can then be detected by a receiver filter. Furthermore, the ratio of the square of energy for the isolated signal and the early signal or the isolated signal and the late signal can be calculated from the cross-correlation, and the ratio helps to search for true synchronization. be able to.

  While preferred embodiments of the invention have been described, other embodiments incorporating this concept may be used, as will be apparent to those skilled in the art. It is therefore felt that these embodiments should not be limited to the disclosed embodiments, but rather should be limited only by the spirit and scope of the appended claims.

1 is a schematic representation of a signal decomposition-resynthesis system. It is a schematic representation of a cascade analyzer. 1 is a schematic representation of a cascade synthesizer. A schematic representation of a tree analyzer. 1 is a schematic representation of a tree synthesizer. It is a block diagram of a signal encryption apparatus. It is a block diagram of the transmission / reception system for transmitting / receiving a high reliability signal. It is a block diagram of a signal encryption transmission / reception system. It is a block diagram of a modem. FIG. 3 is a block diagram of a coded modem. FIG. 3 is a block diagram of a coded tree modem. 1 is a block diagram of a system for sending and receiving compressed signals. 1 is a block diagram of a signal compression system for transmitting and receiving signals over a digital link. It is a block diagram of a modem. 1 is a block diagram of an FM modem according to the present invention. FIG. 12 is a flow diagram of a method for transmitting data using the double sideband variant of the FM modem of FIG. FIG. 12 is a flow diagram of a method for transmitting data using a single sideband variant of the FM modem of FIG. FIG. 14 is a flow diagram of substeps provided by the method of FIGS. 12 and 13. FIG. 12 is a flow diagram of a method for receiving data using the FM modem of FIG. FIG. 16 is a flow diagram of substeps provided by the method of FIG.

Claims (12)

A method for compensating for self-interference of a signal received by a receiver, wherein the receiver comprises a multirate filter bank having a wavelet function as a filter coefficient, and the signal is a wavelet code having a time and a frequency overlap. In a method having a plurality of subbands
Receiving said signal by said receiver by a correlator and analog / digital converter to form a received signal,
To generate a rotation signal set comprising a plurality of bands, the steps of rotating the received signal through the received signal in the multi-rate filter bank,
Filtering the rotated signal set output by the multi-rate filter bank through a second filter bank, the second filter bank being selected between one band and an adjacent band It said filter characterized in that it gives a self-interference amount in the selected metric for each band and having a respective band coefficients respectively calculated from the wavelet inner product of the first value of differential partial change in metric Processing steps ;
Determining a known function of the self-interference amount in the selected measurement criteria for each band,
Weighting the amount of self-interference in the selected metric for each band using the determined known function to generate a correction factor for each band;
To produce an output signal which self-interference is compensated, and subtracting the correction factor for each band outputted by said weighting step from the received signal,
A method comprising the steps of: The selected metric is phase ;
Wherein the step of filtering comprises the step of calculating the respective band coefficients from said wavelet inner product of the first value of the phase shift between the adjacent band and the one band,
The step of determining comprises determining a known function of the amount of self-interference in phase for each band;
The method of claim 1, wherein the weighting step comprises weighting the amount of self-interference in phase for each band using the determined known function . The selected metric is time ;
The filtering process to step includes the step of calculating the respective band coefficients from said wavelet inner product of the first value of the time displacement between the adjacent band and the one band,
The step of determining comprises determining a known function of the amount of self-interference in time for each band;
The method of claim 1, wherein the weighting step comprises weighting the amount of self-interference in time for each band using the determined known function . The method of claim 1, wherein the determining step determines sin (x) as the known function , and wherein x is the amount of self-interference in the selected metric. It said first value of said difference component change in the selected metric, the method according to claim 1, characterized in that it is obtained by measurement. Said first value of said difference component change in the selected metric, the method according to claim 1, characterized in that it is obtained by using an adaptive combiner associated with said receiver. Method for compensating for self-interference introduced in a received signal in a receiver having a plurality of multi- tilt filter banks for analyzing a received signal with wavelets overlapping in time and frequency respectively modulated by data symbols Because
To generate a rotation signal set comprising a plurality of bands components, parsing the received signal using the multirate filter bank,
To generate a self-interference amount in the first metric for each band, from the respective wavelet inner product of a single predetermined value of change in the first metric between adjacent bands component a step of the rotation signal set from the multirate filter bank through a filter bank to filter with the resulting coefficients,
To define the distortion offset, wherein the first metric based on a predefined function of the self-interference amount, the obtained from the results of the filtering process to the step, the first for each band the self-interference amount in metrics, a step of weighting,
Method characterized by comprising the step of subtracting the distortion offset from the received signal to produce an output signal which self-interference is compensated. The first metric is phase;
Wherein the step of filtering comprises one step of filtering the rotation signal set through a filter bank having a coefficient resulting from the respective wavelet inner product in a predetermined value of the phase shift between adjacent bands component ,
The weighting step comprises weighting the self-interference amount in phase for each band based on a predefined function of the self-interference amount in phase. the method of. The first metric is time;
Wherein the step of filtering comprises a step of filtering the rotation signal set through a filter bank having a coefficient resulting from the respective wavelet inner products in one predetermined value of the time delay between adjacent bands component ,
The weighting step includes weighting the self-interference amount in time for each band based on a predefined function of the self-interference amount in time. the method of. The weighting step comprises applying a sine function of x to the self-interference amount in the first metric , wherein x is the self-interference amount in the first metric. 8. A method according to claim 7, characterized in that A receiver for a wavelet transmission system,
A multi-rate filter bank for analyzing a received signal with wavelet-encoded subbands and for generating the rotated signal set comprising a plurality of band components ;
A filter circuit comprising a fixed tap weight filter bank, passing through the fixed tap weight filter bank having coefficients that arise from a wavelet dot product of adjacent subbands as a function of a predefined value of distortion in a first metric. a filter circuit for defining the degree of distortion in the first metric between said for the rotation signal set to filter processing, and the adjacent subbands Te,
A weighting circuit that applies a predetermined weighting function to the defined degree of distortion from the filter circuit to generate a distortion offset , wherein the predetermined weighting function is The weighting circuit being a function of the defined degree of distortion in the first metric between bands ;
A receiver comprising: a subtraction circuit that subtracts the distortion offset from the received signal to generate an output signal in which self-interference is compensated . Wherein the first metric, the receiver of claim 11, wherein is selected from the group consisting of time and phase.

Images